The observations were performed with the VLA operated by the NRAO and with the ATCA operated by the ATNF. Details of the observations are given in Tables 3 and 4. The antenna configurations were chosen to obtain half-power widths of the synthesized beams of about 20 $^{\prime \prime }$ .

In the L band, VLA observations were performed at 1365 MHz ( $\lambda$ 22 cm) and 1665 MHz ( $\lambda$ 18 cm) simultaneously. In each of the VLA C and X bands, the data from two channels were combined (4835 MHz + 4885 MHz and 8435 MHz + 8485 MHz). The four southern galaxies from the VLA sample were observed with hybrid configurations (CnB and DnC) which allow to synthesize a more circular beam. For NGC 1097, the CnB and DnC uv data at the same wavelength were combined.

Several ATCA 750 m configurations were combined at $\lambda$ 5.8 cm (4800 MHz + 5568 MHz) to achieve higher sensitivity and better coverage of the uv plane. For three large galaxies, additional observations with the ATCA 375 m configuration were added. In two observation sessions (1993 May 19 and 1998 January 9), data at 1380 MHz ( $\lambda$ 22 cm) and at 2368 MHz ( $\lambda$ 13 cm) were recorded simultaneously. No significant polarization was detected at $\lambda$ 22 cm. In the session of 1996 October 29, frequencies were set to 2240 MHz ( $\lambda$ 13.4 cm) and 2368 MHz ( $\lambda$ 12.7 cm). In the sessions of 1993 July 26 and September 9, data at 4800 MHz ( $\lambda$ 6.2 cm) and at 8640 MHz ( $\lambda$ 3.5 cm) were recorded simultaneously.

The largest visible structure for full synthesis observations (that requires an observing time in excess of 8 h with the VLA or 12 h with the ATCA) is 15 $^\prime$ at $\lambda$ 18 cm and 22 cm (VLA C or CnB arrays), 5 $^\prime$ at $\lambda$ 6 cm (VLA D or DnC arrays), 3 $^\prime$ at $\lambda$ 3 cm (VLA D or DnC arrays), 3 $^\prime$ at $\lambda$ 13 cm (ATCA 1.5 km arrays), 4 $^\prime$ at $\lambda$ 6 cm (ATCA 750 m arrays) and 8 $^\prime$ at $\lambda$ 6 cm (ATCA 375 m array).

The data were reduced with the standard AIPS and MIRIAD software packages. The maps in Stokes parameters I, Q and U were smoothed to a common resolution of 30 $^{\prime \prime }$ to achieve a higher signal-to-noise ratio. These were combined to maps of total and polarized surface brightness, measured in Jansky per solid angle of the telescope beam ("beam area''), and polarization angle. The positive bias in PI due to noise was corrected by subtraction of a constant value, which is equal to 1.0- $1.4~\times~~$ (rms noise) in the maps of Q and U.

The rms noise in the final maps in I (total intensity) and PI (polarized intensity) is given in Tables 3 and 4. Since the noise in the PI maps has a non-Gaussian distribution (even if Q and U have Gaussian noise) the standard deviation underestimates the noise. Therefore we assume the noise in PI to be the same as that in Q and U. The rms noise in the ATCA maps is typically larger by a factor two in comparison to the VLA maps.

The final maps are displayed in Figs. 5-24, overlayed onto images from the Digitized Sky Surveys. Contours show the total intensity at the wavelength indicated near the upper left corner of each frame, dashes indicate the orientation of the observed E vector of the polarized emission turned by 90 $^\circ$ . These "B vectors'' indicate the orientation of the magnetic field only in case of small Faraday rotation (see below). Due to missing spacings, the VLA maps at $\lambda$ 3 cm and the ATCA maps at $\lambda$ 13 cm do not show the extended emission in full.

We did not attempt to separate the thermal from the nonthermal emission because for most galaxies we have only maps at one or two wavelengths which show the full extended emission. The average thermal fraction in spiral galaxies is only $\simeq$ 10% at $\lambda$ 20 cm (Niklas et al. 1997) which corresponds to $\simeq$ 20% at $\lambda$ 6 cm and $\simeq$ 30% at $\lambda$ 3 cm (assuming a nonthermal spectral index of 0.85).

Figures 25 and 26 show the distribution of polarized intensity PI and the observed B vectors for the galaxies with the strongest polarization. We give the accurate observational wavelengths in the titles, but for ease of reading we will summarize all C-band observations as " $\lambda$ 6 cm'' and all X-band observations as " $\lambda$ 3 cm''. At the shorter wavelengths, $\lambda$ 6 cm and $\lambda$ 3 cm, the B vectors in the figures show the approximate orientation of the magnetic field averaged over the beam. A correction for Faraday rotation, significant at $\lambda\ge13$ cm, was not attempted because of insufficient signal-to-noise ratios of the polarization data at these wavelengths.

In Sect. 4 we quote Faraday rotation measures (RM)between $\lambda$ 22 cm and $\lambda$ 6 cm which, however, can strongly be affected by Faraday depolarization (Sokoloff et al. 1998). RM values between $\lambda$ 6 cm and $\lambda$ 3 cm were computed only for NGC 1097 and NGC 1365 for which the signal-to-noise ratio at $\lambda$ 3 cm is sufficiently high. Correction for Faraday rotation in the Galactic foreground was not attempted.

We note that polarized emission can also be produced by anisotropic turbulent magnetic fields (Laing 1981; Sokoloff et al. 1998; Laing 2002) which can be a result of compression and/or shearing by streaming velocities. These turbulent magnetic fields do not produce any Faraday rotation. The anisotropy of turbulence could be significant in bars. Further Faraday rotation measures with high accuracy and good resolution are required to distinguish between anisotropic turbulent and coherent regular fields in our sample galaxies.

The total and polarized intensities I and PI were integrated in concentric rings (15 $^{\prime \prime }$ wide) defined in each galaxy's plane, using the inclination i and position angle PA given in Tables 1 and 2. The maximum radius for the integration is the outer radius of the ring where I reaches the noise level. The integrated flux density $S_{\lambda }$ is given in Tables 5 and 6. The average degree of polarization $p_{\lambda }$ was obtained from the integrated values of I and PI. The errors in $S_{\lambda }$ and $p_{\lambda }$ include (as a quadratic sum) a 5% uncertainty in the absolute flux calibration and the zero-level uncertainty. In order to determine the zero level, we calculated the average surface brightness in several rings located outside the galaxy image. The rms scatter of these averages was adopted as the zero-level uncertainty.

The variations of the radio flux density and far-infrared flux density between the galaxies of our sample cannot be explained by variations in distance alone. Having scaled the radio flux density at $\lambda$ 6 cm to a common distance of 10 Mpc ( $S_{6~\rm cm}^*$ , a measure of radio luminosity), we define three groups of galaxies (see Fig. 1 and Col. 7 of Tables 5 and 6):

The average total surface bightness $I_{6~\rm cm}$ was computed by dividing $S_{6~\rm cm}$ by the number of beams in the integration area. The results are given in Col. 11 of Tables 5 and 6. The error in $I_{6~\rm cm}$ is dominated by the uncertainty in the integration area which is estimated to be about 25%. A classification based on $I_{6~\rm cm}$ is similar to that based on $S_{6~\rm cm}^*$ .

$\begin{figure} \par\includegraphics[width=8.8cm,clip]{2373fig1a.eps}\end{figure}$

Figure 1: Histogram of the total radio flux densities $S_{6~\rm cm}^*$ , scaled to a distance of 10 Mpc, of the sample galaxies at $\lambda$ 6 cm.

3 Observations and results