Abstract Astyanax mexicanus is a freshwater fish species with blind cave morphs and sighted surface morphs. Like other troglodytic species, independently evolved cave-dwelling A. mexicanus populations share several stereotypic phenotypes, including the expansion of certain sensory systems, as well as the loss of eyes and pigmentation. Here we assess the extent to which there is also parallelism in craniofacial development across cave populations. Since multiple forces may be acting upon variation in the A. mexicanus system, including phylogenetic history, selection and developmental constraint, several outcomes are possible. For example, eye regression may have triggered a conserved series of compensatory developmental events, in which case we would expect to observe highly similar craniofacial phenotypes across cave populations. Selection for cave-specific foraging may also lead to the evolution of a conserved craniofacial phenotype, especially in regions of the head directly associated with feeding. Alternatively, in the absence of a common axis of selection or strong developmental constraints, craniofacial shape may evolve under neutral processes such as gene flow, drift, and bottlenecking, in which case patterns of variation should reflect the evolutionary history of A. mexicanus. Our results found that cave-adapted populations do share certain anatomical features; however, they generally did not support the hypothesis of a conserved craniofacial phenotype across caves, as nearly every pairwise comparison was statistically significant, with greater effect-sizes noted between more distantly related cave populations with little gene flow. A similar pattern was observed for developmental trajectories. We also found that morphological disparity was lower amongst all three cave populations versus surface fish, suggesting eye loss is not associated with increased variation, which would be consistent with a release of developmental constraint. Instead, this pattern reflects the relatively low genetic diversity within cave populations. Finally, magnitudes of craniofacial integration were found to be similar among all groups, meaning that coordinated development among anatomical units is robust to eye loss in A. mexicanus. We conclude that in contrast to many conserved phenotypes across cave populations, global craniofacial shape is more variable, and patterns of shape variation are more in line with population structure than developmental architecture or selection.