We use extensive multi-wavelength photometric data from the Great Observatories Origins Deep Survey (GOODS) to estimate photometric redshifts for a sample of 434 galaxies with spectroscopic redshifts in the Chandra Deep Field South. Using the Bayesian method, which incorporates redshift/magnitude priors, we estimate photometric redshifts for galaxies in the range 18 < R (AB) < 25.5, giving an rms scatter of 0.11. The outlier fraction is < 10%, with the outlier-clipped rms being 0.047. We examine the accuracy of photometric redshifts for several, special sub--classes of objects. The results for extremely red objects are more accurate than those for the sample as a whole, with rms of 0.051 and very few outliers (3%). Photometric redshifts for active galaxies, identified from their X-ray emission, have a dispersion of 0.104, with 10% outlier fraction, similar to that for normal galaxies. Employing a redshift/magnitude prior in this process seems to be crucial in improving the agreement between photometric and spectroscopic redshifts.