In a series of three papers, we investigate the current status of published metallicities for open clusters that were derived from a variety of photometric and spectroscopic methods. The current article focuses on spectroscopic methods. The aim is to compile a comprehensive set of clusters with the most reliable metallicities from high-resolution spectroscopic studies. This set of metallicities will be the basis for a calibration of metallicities from different methods. The literature was searched for [Fe/H] estimates of individual member stars of open clusters based on the analysis of high-resolution spectra. For comparison, we also compiled [Fe/H] estimates based on spectra with low and intermediate resolution. At medium and high resolution, we found that differences in the analysis methods have a stronger effect on metallicity than quality differences in the observations. We retained only highly probable cluster members and introduced a restriction on atmospheric parameters. We combined 641 individual metallicity values for 458 stars in 78 open clusters from 86 publications to form our final set of high-quality cluster metallicities. The photometric metallicities discussed in the first paper of this series are systematically lower than the spectroscopic ones by about 0.1 dex, and the differences show a scatter of about 0.2 dex. In a preliminary comparison of our spectroscopic sample with models of Galactic chemical evolution, none of the models predicts the observed radial metallicity gradient. Photometric metallicities show a large intrinsic dispersion, while the more accurate spectroscopic sample presented in this paper comprises fewer than half the number of clusters. Only a sophisticated combination of all available photometric and spectroscopic data will allow us to trace the metallicity distribution in the Galactic disk on a local and global scale.< Réduire