Background: Avian plumage is ideal for investigating phenotypic convergence because of repeated evolution of the same within-feather patterns. In birds, there are three major types of regular patterns within feathers: scales, bars and spots. Existing models of within-feather pattern development suggest that scales have the simplest developmental mechanism, bars require more stringent regulation than scales, and spots have the strictest developmental parameters. We hypothesized that increasing stringency in the mechanism of pattern formation predicts the evolutionary trajectory of patterns, and hence scales should evolve first, followed by bars and finally spots. Here, using Bayesian phylogenetic modeling we reconstructed pattern evolution in the most spectacularly patterned avian clades – aquatic waterfowl (Anseriformes) and terrestrial gamebirds (Galliformes). Results: Our analyses suggest that the ancestral state of plumage is an absence of patterns, but with some variability. Independent analyses of seven feather patches reveal that spots evolve after bars and scales. However, both scales and bars evolve frequently from an absence of patterns, contradicting our predictions. Over the whole body, many constraints are conserved from the level of patches, for example the largest number of steps from the ancestral state was required for spots to evolve. Conclusions: Overall there was remarkable similarity in the inferred evolutionary trajectories of plumage pattern evolution in Galliformes and Anseriformes, suggesting that developmental constraint is similar in these two orders, despite large ecological differences. These evolutionary transitions are largely congruent with a reaction–diffusion based model of pattern formation, but the evolution of bars from an unpatterned ancestor is more common than expected. Our study highlights the promise of testing models of development using comparative methods.