The observation that phenotypic convergence and genetic convergence are widespread in nature implies that evolution is at least somewhat predictable. But to what extent and under what circumstances? In other words, how predictable is evolutionary predictability? Answering this question requires going beyond documenting examples of repeated evolution to actually quantifying predictability at different hierarchical levels. At present, few such studies exist. In this issue of Molecular Ecology, Chaturvedi et al. () quantify the predictability of genomewide changes that accompany shifts to an introduced host plant (alfalfa) in populations of the Melissa blue butterfly (Lycaeides melissa). They evaluate predictability in two contexts: (i) overlap in host-associated loci among populations that have independently colonized alfalfa, and (ii) overlap between host-associated loci in nature and loci associated with host performance in laboratory experiments. Overall, they find that the genomic changes that accompany host shifts in this system are indeed somewhat predictable. However, the degree of predictability depends on the type of comparison (among natural populations vs. between natural and experimental populations), type of convergence (specific genomic locations vs. direction of allele frequency change), geographic scale (rangewide vs. specific population pairs) and location in the genome (autosomes vs. sex chromosomes). Together with a handful of comparable data sets, Chaturvedi et al.'s () work suggests that the relative contribution of stochastic and deterministic processes to genomewide responses to novel selection pressures may be highly variable, but possibly predictably so.