BACKGROUND: A surprising amount of developmental variation has been observed for otherwise highly conserved features, a phenomenon known as developmental system drift. Either stochastic processes (e.g., drift and absence of selection-independent constraints) or deterministic processes (e.g., selection or constraints) could be the predominate mechanism for the evolution of such variation. We tested whether evolutionary patterns of change were unbiased or biased, as predicted by the stochastic or deterministic hypotheses, respectively. As a model, we used the nematode vulva, a highly conserved, essential organ, the development of which has been intensively studied in the model systems Caenorhabditis elegans and Pristionchus pacificus. RESULTS: For 51 rhabditid species, we analyzed more than 40 characteristics of vulva development, including cell fates, fate induction, cell competence, division patterns, morphogenesis, and related aspects of gonad development. We then defined individual characters and plotted their evolution on a phylogeny inferred for 65 species from three nuclear gene sequences. This taxon-dense phylogeny provides for the first time a highly resolved picture of rhabditid evolution and allows the reconstruction of the number and directionality of changes in the vulva development characters. We found an astonishing amount of variation and an even larger number of evolutionary changes, suggesting a high degree of homoplasy (convergences and reversals). Surprisingly, only two characters showed unbiased evolution. Evolution of all other characters was biased. CONCLUSIONS: We propose that developmental evolution is primarily governed by selection and/or selection-independent constraints, not stochastic processes such as drift in unconstrained phenotypic space.