While the spectra obtained in this survey lack the resolution and signal-to-noise ratio needed to produce an accurate spectral classification of each target, they are nevertheless useful for assigning them to one of several broad categories, as well as to guide future detailed investigations of particular objects.

Both the existence of lightly obscured members of Cygnus OB2 and the absence of any significant nearby intervening cloud over the surveyed area, which would stand out as a decrease in stellar density in visible-light images of the region, suggest that most of the extinction on the objects plotted in the 2MASS color-color and color-magnitude diagrams is produced at a distance similar to that of Cygnus OB2 or greater. Our selection of targets that are both very red and bright in the infrared thus ensures that they are intrinsically bright, thus implying that the vast majority should fall within one of two categories: hot luminous young stars likely to belong to the association, and cool red giants and supergiants. The distinction between both groups can be established easily from our spectra, mainly based on the presence or absence of CO bands in absorption longwards of 2.29 $\mu$ m. Even for early K-type giants with less prominent CO bands that may not be clearly seen in some of our noisier spectra, the shape of the continuum still allows a clear distinction between them and the much hotter early-type stars whose infrared spectral energy distribution is well matched by a reddened Rayleigh-Jeans tail. Of the cool stars group, only red supergiants might belong to the association on the basis of their ages and evolutionary status, a possibility that we will consider below.

At the resolution and signal-to-noise of the spectra under consideration here O- and early B-type stars with a purely absorption-line spectrum should look featureless, the noise dominating over the depth of the strongest features due to H and He. Lacking more information from the current material, we thus classify as candidate early-type stars in our sample those that appear as a featureless continuum.

We have thus identified 77 early-type candidates in our sample. Of these, 68 belong to the "blue'' group (H-K) < 0.5, and many of them have therefore been included in surveys in the visible. Accurate spectral classifications are available in the literature for 31 of them, mostly from the works of Massey & Thompson (1991) and Hanson et al. (1996). Most of these stars are O-type (24). The remaining 7 have been classified in the literature as B0V (2), B1.5V (1), B1II (1), or simply B (3). Detailed studies on some of the most luminous among these objects can be found in Herrero et al. (1999). Confirmed contamination among early-type candidates by non-OB stars in the blue group is found to be small, as expected from the small distance of the sample to the reddening band of early-type objects in the infrared color-color diagram: one star in this group has also a published classification as F5 and is thus foreground, and four other stars that were included in the blue sample on the basis of their infrared colors turned out to display clearly visible hydrogen absorption lines classifying them as foreground A-type stars, confirmed in three cases by published spectral classifications in the visible.

The remaining 37 stars having featureless spectra in the group of (H-K) < 0.5, as well as the 9 stars with featureless spectra in the (H-K) > 0.5 group, are new candidate members of the association and are listed in Table 1. None of the latter appears to be associated to sources seen at other wavelengths, with the exception of A1 whose position is consistent with that of the radio-continuum source MITG J2033+4041 (Griffith et al. 1991) appearing in the near-infrared as a small, faint HII region around this star (Fig. 6). Given the limited positional accuracy of the MITG catalog ( $\sim$ 40'' rms), the K-band image confirms the association of the star with the HII region. Some of the stars in Table 1 had been already included in the lists of Reddish et al. (1966), Massey & Thompson (1991) and Parthasarathy & Jain (1995), but spectral classification was not available. At least two of these new members seem to be the counterparts of stellar soft X-ray sources detected by the Einstein satellite, a property that is common among the brigther members of Cygnus OB2 (Waldron et al. 1998).

**Table 1:** Candidate new early-type members of Cygnus OB2 with featureless low-resolution spectra.
Number	$\alpha (2000)$	$\delta(2000)$	K	(J-H)	(H-K)	remarks
A1	20 33 11.7	40 41 54.4	9.987	1.83	1.637	possibly associated to radio-continuum source [1]
A2	20 32 51.4	41 8 40.0	9.696	1.534	1.285
A3	20 31 33.8	41 19 15.4	9.932	1.29	1.066
A4	20 31 36.3	41 22 3.3	5.865	1.19	0.661
A5	20 35 9.8	41 35 29.7	7.885	1.063	0.611
A6	20 32 8.3	40 25 7.0	9.104	1.073	0.574
A7	20 34 43.	40 29 30.3	9.042	1.062	0.568
A8	20 33 41.6	41 47 57.1	6.792	1.202	0.565
A9	20 35 32.7	41 20 55.2	9.056	1.005	0.519
A10	20 34 55.1	40 34 44.0	8.261	0.864	0.454
A11	20 32 31.5	41 14 8.0	6.637	0.745	0.442	possibly associated to X-ray source 1E203043+4103.9 [2]
A12	20 33 38.2	40 41 6.0	5.723	0.776	0.429
A13	20 33 1.2	40 32 33.0	8.491	0.755	0.408
A14	20 31 18.9	42 2 56.0	8.464	0.725	0.406
A15	20 31 36.8	40 59 9.0	6.805	0.733	0.406
A16	20 34 36.9	40 41 2.0	8.789	0.74	0.388
A17	20 32 35.3	41 14 45.0	8.731	0.697	0.383	# 886 [3]
A18	20 30 7.8	41 23 50.0	8.347	0.685	0.38
A19	20 31 25.9	41 16 2.0	8.504	0.674	0.38	# 597 [4]
A20	20 33 2.9	40 47 25.0	6.248	0.648	0.371	possibly associated to X-ray source 2E2031.2+4037 [5]
A21	20 29 34.8	41 20 8.0	8.483	0.647	0.353
A22	20 33 11.2	40 42 33.0	8.076	0.684	0.353
A23	20 30 39.7	41 8 48.0	5.963	0.633	0.346
A24	20 34 44.1	40 51 58.0	7.461	0.625	0.345
A25	20 32 38.4	40 40 44.0	7.358	0.671	0.341
A26	20 30 57.7	41 9 57.0	8.187	0.643	0.33
A27	20 34 44.7	40 51 46.0	5.745	0.638	0.329
A28	20 34 16.	41 2 19.0	8.526	0.594	0.322
A29	20 34 56.	40 38 18.0	6.541	0.594	0.318
A30	20 31 22.1	41 12 3.0	8.61	0.489	0.318
A31	20 32 39.5	40 52 47.0	7.976	0.633	0.312	KMJ 1718 [6]
A32	20 32 30.3	40 34 33.0	7.038	0.554	0.312
A33	20 32 34.9	40 52 39.0	8.598	0.565	0.305
A34	20 31 36.9	42 1 21.0	6.653	0.378	0.305
A35	20 30 55.5	40 54 54.0	8.471	0.506	0.305
A36	20 34 58.7	41 36 17.0	6.36	0.544	0.302	# 357 [7]
A37	20 36 4.5	40 56 13.0	7.675	0.614	0.292
A38	20 32 34.8	40 56 17.0	8.564	0.553	0.292	KMJ 1315 [6]
A39	20 32 27.3	40 55 18.0	7.857	0.492	0.285	KMJ 1834 [6]
A40	20 35 13.6	40 55 25.0	8.391	0.601	0.284
A41	20 31 8.3	42 2 42.0	7.029	0.517	0.269
A42	20 29 57.	41 9 53.0	8.444	0.439	0.266
A43	20 32 38.5	41 25 13.0	7.699	0.306	0.222
A44	20 31 46.	40 43 24.0	8.262	0.46	0.211
A45	20 29 46.6	41 5 8.0	8.455	0.425	0.192
A46	20 31 0.1	40 49 49.0	7.838	0.39	0.178

4.2 Stars with Br $\mathsf{\gamma}$ in emission

Our survey also revealed 20 stars with discernible emission in Br $\gamma$ (2.166 $\mu$ m) and occasionally in other lines as well. Besides these 20 stars, our selection criteria also included the three Wolf-Rayet stars WR 144, WR 146, and V1923 Cyg. The complete list is given in Table 2, and the full set of spectra appears in Fig. 7. The equivalent widths of a number of features is listed in Table 3. Only 4 of these emission-line stars have published spectra in the literature: these are Cyg OB2 # 360 (B1.5III...; Massey & Thompson 1991), VI Cyg 9 (O5Iab:e; Herrero et al. 1999), VI Cyg 11 (O6e; Herrero et al. 1999), and Cyg OB2 # 708 (B1.5V and Herbig Ae/Be; Thé et al. 1994). Three other stars with Br $\gamma$ emission have been included in previous lists of Cygnus OB2 members but without published spectral types, including V2188 Cyg (Pigulski & Kolaczkowski 1998). We also note that three of the reddest objects are probably associated with IRAS point sources.

**Table 2:** Observed stars with Br $\gamma$ in emission.
Number	$\alpha (2000)$	$\delta(2000)$	K	(J-H)	(H-K)	remarks
B1	20 36 7.5	41 40 9.3	8.893	4.229	3.1	possibly associated to IRAS 20343+4129 [8]
B2	20 34 13.3	41 8 14.0	8.534	2.977	2.455
B3	20 34 30.9	41 30 39.1	7.288	3.439	2.413	possibly associated to IRAS 20327+4120
B4	20 34 47.	40 54 48.6	8.887	2.264	1.653
B5	20 34 2.9	41 10 40.9	9.982	2.036	1.432
B6	20 32 11.3	40 40 34.0	8.586	1.701	1.427	possibly associated to IRAS 20303+4030
B7	20 34 13.4	41 1 57.8	9.893	1.96	1.397
B8	20 33 25.6	41 8 31.5	9.667	1.579	1.307
B9	20 35 16.3	41 12 36.2	9.177	1.145	0.736
B10	20 33 27.1	41 35 58.5	8.177	0.777	0.67	# 285 [2]
B11	20 33 18.5	41 15 35.2	7.952	0.862	0.623	V2188 Cyg [9]; [3]
B12	20 32 30.9	41 10 0.6	8.692	0.852	0.598
B13	20 33 34.3	41 18 11.4	7.826	0.723	0.585	# 708; B1.5V Herbig Ae/Be [12]
B14	20 34 5.6	40 52 19.6	8.91	0.852	0.532
B15	20 33 18.7	40 59 37.0	8.127	0.657	0.457
B16	20 34 43.5	41 29 4.0	7.718	0.516	0.409	# 360, B1.5III [2]
B17	20 30 27.3	41 13 25.0	6.443	0.804	0.406
B18	20 34 57.8	41 43 54.0	7.42	0.672	0.365
B19	20 33 10.7	41 15 8.0	5.556	0.611	0.332	VI Cyg 9: O5Iab:e... [11]
B20	20 34 8.5	41 36 59.0	5.993	0.443	0.236	VI Cyg 11: O6e [11]

Notes: same as in Table 1, plus: [8] Miralles et al. (1994); [9] Pigulski & Kolaczkowski (1998); [10] Thé et al. (1994); [11] Herrero et al. (1999).

Table 3: Equivalent widths of emission lines (in Å).
Number Br $\gamma$ Br16 Br15 Br14 Br13 Br12 Br11 Br10 HeI HeI

2.166 1.555 1.569 1.588 1.611 1.641 1.681 1.736 1.700 2.058 2.225

B1 3.3 7.3

B2 6.3

B3 17.2 6.8 13.4 16.6

B4 10.6

B5 8.9

B6 10.3 9.9

B7 12.2

B8 10.9

B9 27.0 2.7 4.8 12.6 15.3 16.6 16.3 19.4

B10 29.6 5.7 8.7 14.0 17.3 17.1 19.2 21.3 8.4

B11 22.2 3.8 6.7 11.7 9.1 14.9 15.9 14.8

B12 22.0 6.5 7.5 9.8 10.0 9.4 3.1

B13 13.5 1.6 3.8 7.1 8.9 10.1 12.4 12.4 4.2

B14 23.2 10.6 18.8

B15 9.9 4.7 7.6 6.3 12.6 16.1 4.4

B16 23.7 11.3 13.2 13.7 15.3 15.1 4.4

B17 13.4 6.9 7.0

B18 17.1

B19 6.2

B20 8.2

Despite the common feature of having Br $\gamma$ in emission, Fig. 7 shows a variety of spectral morphologies found among these objects. The features in their spectra are characteristic of those displayed in late stages of massive star evolution (WNL, Of, Ofpe/WN9, Be, B[e], and LBVs), but assigning the different classes of emission-line spectra that we obtain to any of those stages is not possible. In their extensive review on the near-infrared properties of massive stars in transition, Morris et al. (1996) stress the difficulty of assigning unambiguous spectroscopic classification criteria to separate among the different groups established from observations in the visible, finding evidence for a close relationship among them. A study of our objects to a comparable level of detail requires better quality spectra than currently available and is clearly beyond the scope of this paper, but the results plotted in Fig. 7 already provide a glimpse at the diversity found also in our sample. For example, it is interesting to note the differences in the Brackett decrement between the group formed by B9, B10, B11, B12, B13, B15, and B16 on one side, and B14 and B18 (at least; the H-band spectra of B3, B7, and B8 are too faint for a reliable measurement of the higher Brackett lines) on the other. The systematic difference is reminiscent of that found by Morris et al. between Ofpe/WN9 stars and LBVs. The HeI line at 2.058 $\mu$ m is clearly seen in B10 and B14, and is possibly present also in B9, B12, B13, and B15, while the HeI line at 1.700 $\mu$ m is seen in B16 and B17. Wide variations of this line have been measured by Morris et al. within each of the groups of evolved massive stars mentioned above.

K-band images obtained for one of our reddest targets, B3, shows a compact, asymmetric nebula roughly $30\arcsec$ across around it (Fig. 8). Both the mid-infrared point source IRAS 20327+4120 and the water maser emission detected near that position by Palla et al. (1991) are likely to arise in this nebulosity. We finally note the reddest object in our sample, B1, whose spectrum is featureless with the exception of a clearly discernible emission line at 2.225 $\mu$ m. Weak Br $\gamma$ emission is also probably present. The stronger emission might be due to [FeII], sometimes observed in the spectra of LBVs (see the spectra of WRA 751 in Morris et al. 1996) but not as a dominant or unique emission feature. B1 is probably associated to IRAS 20343+4129, which has been detected as a faint, unresolved thermal radiosource at 6 cm by Miralles et al. (1994). Its mid-infrared properties and the association with dense gas traced by NH₃ (Miralles et al. 1994) agree with it being an ultracompact HII region. However, the brightness at K argues against B1 and IRAS 20343+4129 being the same object, as the central star of the HII region should be deeply embedded and highly obscured even in the near-infrared. Clearly, the nature of B1 and its relationship to the far-infrared and the radiocontinuum sources deserve further study. The spectral type B2 assigned by Miralles et al. is derived under the assumption that all the ionizing photons of the stars are used in ionizing the nebula neglecting the presence of dust, and is thus a lower limit.

4.3 Stars with CO in emission

While none of the Br $\gamma$ -emitting stars in our sample displays detectable CO overtone emission that is sometimes observed in B[e], Be, and LBV stars, three objects in our sample do show CO in emission, as shown in Fig. 9 and listed in Table 4. These objects are reddened by widely different amounts and only one of them, C3 in Table 3, is associated with an object identified in the visible, Cyg OB2 #1071.

**Table 4:** Observed stars with CO in emission.
Number	$\alpha (2000)$	$\delta(2000)$	K	(J-H)	(H-K)	remarks
C1	20 34 43.3	40 53 13.7	9.342	2.954	2.146	probably associated to IRAS 20328+4042
						and radio continuum source [12]
C2	20 32 31.2	41 14 44.1	9.814	1.409	1.186
C3	20 34 10.4	41 2 48.0	9.417	0.897	0.862	[3]

Notes: same as in Table 1, plus: [12] Wendker et al. 1991, A&A, 241, 551.

The spectrum of C1 displays strong emission at 2.12 $\mu$ m, besides the CO emission. Intense emission near that wavelength is often seen in the spectra of evolved massive stars due to HeI. However, due to the absence of other helium lines and of Br $\gamma$ , together with the simultaneous existence of strong CO emission, we are inclined to attribute the emission at that wavelength in C1 to H₂ instead, with both H₂ and CO indicating the presence of a massive circumstellar envelope or disk. The presence of important amounts of molecular gas and dust around this object appears to be confirmed by the likely association of C1 with IRAS 20328+4042, although the offset between the coordinates of both sources, 29'', is only marginally within the IRAS error ellipse. This offset may be explained by the existence of small extended emission around C1, clearly visible in the K-band image of the field presented in Fig. 10. The arc-shaped nebulosity around C1 is suggestive of dynamical interaction with the surrounding medium. Faint resolved emission in the 21 cm continuum is also observed at the position of C1 (Wendker et al. 1991).

The other two objects in our survey showing CO overtone emission do not display any other emission features in our spectra nor are associated with either radio continuum or mid-infrared sources. Their relatively blue spectral energy distributions suggest that they are emerged objects whose CO emission may be due to a moderately massive circumstellar disk.

4.4 Possible red supergiants

The considerable number of evolved massive stars that we find in our sample suggests that red supergiants may also be present in Cygnus OB2. Our selection criteria described in Sect. 3 is not designed to detect these stars, but rather tends to actually exclude them, as red supergiants should lie along the upper reddening band in the color-color diagram. However, the fact that for (H-K) > 0.5 we included many stars already in that band as demonstrated by the large number of stars with cool near-infrared spectra that we encounter, has prompted us to have a detailed look at those searching for possible supergiants belonging to the association.

The K-band spectra of late-type supergiants are characterized by strong CO absorption bands whose depth is matched only by late-type Miras (Lançon & Wood 2000). However, unlike the latter, late-M supergiants do not display the prominent water vapour bands that form in the extended atmospheres of Miras. We have thus used the approach proposed by Comerón et al. (in preparation) to select late-M supergiants based on the strengths of these molecular features, based on the following spectrophotometric bands:

where use is made of the Rieke & Lebofsky (1985) extinction law and an intrinsic color (H-K)₀ = 0.30 is assumed for M supergiants (Tokunaga 2000). The following indices measuring the strength of the CO and H₂O features are then approximately reddening-independent:

Let us finally stress the fact already mentioned at the beginning that our failure to identify any red supergiants in Cygnus OB2 does not imply that such stars are absent from the association, since our color-based selection criterion is biased against such objects.

4.5 Color-color and color-magnitude diagrams of candidate members of Cygnus OB2

Figures 11 and 12 show the positions occupied in color-color and color-magnitude diagrams of all spectroscopically observed members of the association. In Fig. 11, nearly all the early-type stars with H-K < 0.5 are well aligned in a band parallel to the reddening vector but offset by $\Delta (J-H) \simeq 0.1$ with respect to it. This may be indicative of a small incorrectness in the adopted intrinsic color calibration, as many of these stars are known to have earlier spectral types and hence colors bluer than those of a B0 star, where we have placed the origin of the reddening vector. Most of the newly found early-type stars with featureless spectra but redder colors are found in the extension of the same line.

$\begin{figure} \par\includegraphics[width=7cm,clip]{MS1699f11.eps} \end{figure}$

Figure 11: Color-color diagram of all the objects observed. Small asterisks, blue early-type stars (H-K < 0.5); circles, reddened early-type stars (H-K > 0.5); squares, stars with Br $\gamma$ in emission; triangles, stars with CO in emission; large asterisks, WR stars. The dashed line is the locus occupied by reddened B0 stars with intrinsic colors J-H = -0.12, H-K = -0.04, corresponding to a B0 star according to Tokunaga (2000); the end of the line corresponds to an extinction of A_V = 40.

The vast majority of the emission-line stars, as well as the stars with CO emission, the three WR stars, and three of the stars classified as having non-emission, early-type spectra, are found to the right of the reddening vector, indicating the existence of significant amounts of dust emission in their surroundings. The range of extinctions is difficult to assess precisely, as this would need the disentangling of the contribution of the circumstellar emission and the foreground extinction to the position in the diagram. However, the fact that the reddest objects tend to display spectroscopic signatures of either strong winds or excitation of molecular gas suggests that the origin of the red colors is largely contributed by extinction produced by dust around the star, either in a stellar wind-blown nebula or in the material left over from the formation of the star.

Figure 12 shows that the three reddest emission-line stars that we observed have K magnitudes placing them among the brightest members of Cygnus OB2 if they were dereddened along the extinction vector. This is actually an upper limit to their intrinsic brightness, because as we have argued considerable contribution to the flux by circumstellar material may be expected. However, Fig. 13 shows that this is still true when the H, (J-H) diagram, which should be less affected by circumstellar emission, is considered. Sources B1, B2, and B3 thus appear as especially interesting objects for further studies of the most massive and luminous objects in Cygnus OB2, particularly in view to clarify their relationship with their associated IRAS sources in the cases of B1 and B3.

4.6 Spatial distribution

The nearly extinction-free view of the Cygnus OB2 region provided by our infrared-selected sample yields a much more unbiased picture of the extent of the association than available from a census of members identified in the visible. As pointed out by Knödlseder (2000), the true shape of Cygnus OB2 emerging from infrared star counts seems to be much rounder than that derived from visible surveys, the latter being dominated by an irregular distribution of the extinction in front of the association.

The positions of all the known and new candidate members of Cygnus OB2 that we observed are plotted in Fig. 14. Since our sample is restricted to candidate O-type stars and thus has a much brighter limit than the sample considered by Knödlseder, our distribution of points in Fig. 14 is far more sparse than that leading to the map shown in Fig. 5 of Knödlseder. Our restriction to a radius of one degree from the center of the association also may leave outside some of the outermost regions of Cygnus OB2. However, our results already seem to confirm the deviation from circular symmetry found by Knödlseder. We note in particular the scarcity of members towards the East and the North of the center of the association, while the larger extension towards the Southwestern quadrant, most probably reaching beyond the boundaries of the one-degree radius selected in our study, is obvious.

A comparison among the spatial distributions of the different types of massive stars considered here shows some possible intriguing differences. The concentration of bluer early-type stars (H-K < 0.5) towards the Southwestern quadrant is more pronounced, as may be expected as a consequence of the somewhat higher extinction towards the East and Northeast. However, no obvious asymmetry is seen in the distribution of emission-line stars, which are slightly more represented in the Northeastern quadrant than in the Southwestern one. We should refrain from making a definitive statement on the reality of this difference, as the low number of emission-line stars makes the statistical significance of this difference rather limited. It may be tempting to speculate that the scarcity of emission-line (e.g. evolved) massive stars with respect to normal O and early B stars in the Southwestern quadrant of the association may reflect a younger average age of that region, perhaps as consequence of the progression of star formation in that direction. However, the small number of stars on which this conjecture is based makes it only tentative, and we mention it here just as a possible direction for future observations.

4.7 The O-star contents of Cygnus OB2

The rich contents of Cygnus OB2 in O-type stars, or in stars having evolved from a O-type progenitor, has been known since the earliest studies on this association, and has been quantified more recently by several authors (Massey & Thompson 1991; Massey et al. 1995; Parthasarathy & Jain 1995; Knödlseder 2000) who have investigated the upper end of its mass function on the basis of members identified either in the visible or the infrared. An important driver of recent investigations on Cygnus OB2 has been the recognition of its importance in linking the emission of ²⁶Al observed in the Cygnus region to nucleosynthesis processes in the interior of the massive stars of the association (del Río et al. 1996; Cerviño et al. 2000; Knödlseder 1999), within the broader goal of understanding the contribution of OB associations to the chemical enrichment processes in galactic disks. With a different scope, Comerón et al. (1998) have considered the energetic activity of the most massive stars in Cygnus OB2 as a possible trigger of star formation in its neighbourhood, reflected nowadays in the peculiar motions of stars in other associations surrounding it.

It is thus interesting to estimate the O-star contents of Cygnus OB2 in the light of our results, and to compare it to the most recent estimate of Knödlseder (2000), as both should be nearly free from extinction biases. Let us start by considering the 77 early-type stars that we have observed: of these, 31 have been spectroscopically classified in the visible and 24 have been found to be O-type, as described in Sect. 4.1, the other 7 belonging to B type. Assuming that the same ratio holds for the entire set of 77 stars, we thus estimate 60 O-type stars among those.

As described in Sect. 4.2, the stars with Br $\gamma$ in emission are expected to represent different stages of the evoution of the most massive stars. Assuming that all these stars started their lives as O stars, this adds 20 more stars. This may not be the case for all these objects however, especially for the stars closer to the B0 limit in Fig. 12, given the infrared excess commonly displayed by these objects: Fig. 13 shows that one of these stars is slightly fainter than the B0 limit in the H, (J-H) diagram, and five more are very close to it, so the actual number of evolved massive stars with O-type progenitors may be somewhere between 14 and 20. Other stars that obviously have evolved from an O-type phase are the three WR stars, thus raising the census to 76-83 objects. We note that the spectral characteristics of these 23 emission-line stars, of which 16 have been identified as such for the first time in the present work, indicate strong mass loss rates, thus being important for the present-day mechanical energy input in the association.

We finally consider the three stars with CO emission, in which circumstellar material is also expected to contribute to the luminosity. Two of them are close to the B0 boundary, and fall well below it in the H, (J-H) diagram of Fig. 13; their brightness at K is thus likely to be largely due to circumstellar emission, and we consider it unlikely that they are related to O-type stars. While the circumstellar contribution to the flux at K is also important for the third one, C1, its high luminosity also in H leads us to include it in our census of O-type stars.

We thus estimate that we have identified between 77 and 84 O-type stars or closely related objects in the association. This may not be a complete census however, mainly due to two reasons. First, as Fig. 5 suggests, our observations do not completely sample the band of lightly reddened (H-K < 0.5) early-type stars; there are 23 objects along this band for which the available time did not allow us to obtain spectra. Fortunately, spectral classifications are found in the literature for 14 of them, mostly in Massey & Thompson (1991) but also from Herrero et al. (1999) and Walborn & Howarth (2000, for the remarkable O3Iab star VI Cyg 7). Of these, 10 are classified as O-type stars, one is a B1Ib star, and three are classified as F- and G-type and thus likely foreground. Assuming that similar ratios apply for the other 9 stars without spectral classification, this makes us estimate that there are about 16 stars in thus group. A second source of incompleteness may be the extension of the association beyond our one-degree-radius limit, especially in the Southeastern direction. The stellar density contours presented by Knödlseder (2000) do not show any significant enhancement in those regions. However, we should keep in mind that even a low density halo might contain a non-negligible fraction of the members of Cygnus OB2, given the quadratic growth of projected area with radius.

Our overall estimate on the total number of O-type or related stars in Cygnus OB2 is thus 90-100, slightly below Knödlseder's estimate but well compatible within uncertainties. Of these, approximately 25% display emission lines in our spectra (including possible Br $\gamma$ -emitting stars among the non-observed part of the bluer sample). The confirmation of Knödlseder's estimate at the highest masses support his conclusions on the total stellar contents and reaffirms Cygnus OB2 as one of the most massive stellar associations in our Galaxy, perhaps comparable to a young globular cluster as claimed by that author.

4 Results