The distribution of genetic variation within and among populations is commonly used to infer their demographic and evolutionary histories. This endeavour has the potential to benefit substantially from high-throughput next-generation sequencing technologies through a rapid increase in the amount of data available and a corresponding increase in the precision of parameter estimation. Here we report the results of a phylogeographic study of the North American butterfly genus Lycaeides using 454 sequence data. This study serves the dual purpose of demonstrating novel molecular and analytical methods for population genetic analyses with 454 sequence data and expanding our knowledge of the phylogeographic history of Lycaeides. We obtained 341,045 sequence reads from 12 populations that we were able to assemble into 15,262 contigs (most of which were variable), representing one of the largest population genetic data sets for a non-model organism to date. We examined patterns of genetic variation using a hierarchical Bayesian analysis of molecular variance model, which provides precise estimates of genome-level phi(ST) while appropriately modelling uncertainty in locus-specific phi(ST). We found that approximately 36% of sequence variation was partitioned among populations, suggesting historical or current isolation among the sampled populations. Estimates of pairwise genome-level phi(ST) were largely consistent with a previous phylogeographic model for Lycaeides, suggesting fragmentation into two to three refugia during Pleistocene glacial cycles followed by post-Pleistocene range expansion and secondary contact leading to introgressive hybridization. This study demonstrates the potential of using genome-level data to better understand the phylogeographic history of populations.