Evidence of phenotypic parallelism is often used to infer the deterministic role played by natural selection. However, variation in the extent or direction of divergence is often evident among independent evolutionary replicates, raising the following question: just how parallel, overall, is parallel evolution? We answer this question through a comparative analysis of studies of fishes, a taxon where parallel evolution has been much discussed. We first ask how much of the among-population variance in phenotypic traits can be explained by different "environment" types, such as high predation versus low predation or benthic versus limnetic. We then use phenotypic change vector analysis to quantify variation in the direction (vector angles) and magnitude (vector lengths) of environment-associated divergence. All analyses show high variation in the extent of parallelism-from very high to very low, along with everything in between-highlighting the importance of quantifying parallelism rather than just asserting its presence. Interestingly, instances of low extents of parallelism represent important components of divergence in many cases, promising considerable opportunities for inferences about the factors shaping phenotypic divergence.