The presence of additional hox clusters in the zebrafish has led to the hypothesis that there was a whole genome duplication at the origin of modern fish. To investigate the generality of this assumption, we analyzed all available actinopterygian fish gene families, and sequenced nuclear receptors from diverse teleost fish. The origin and timing of duplications was systematically determined by phylogenetic analysis. More genes are indeed found in zebrafish than in mouse. This abundance is shared by all major groups of euteleost fish, but not by eels. Phylogenetic analysis shows that it may result from frequent independent duplications, rather than from an ancestral genome duplication. We predict two zebrafish paralogs for most mouse or human genes, thus expressing a note of caution in functional comparison of fish and mammalian genomes. Redundancy appears to be the rule in fish developmental genetics. Finally, our results imply that the outcome of genome projects cannot be extrapolated easily between fish species.