We analyze sequences from two mitochondrial genes, cytochrome b (cyt b) and 12S rRNA (12S), for a group of sigmodontine rodents among which phylogenetic relationships are well understood based on concordance of morphological, chromosomal, allozyme, and other DNA data sets. Because these two genes are physically linked on the nonrecombining mitochondrial genome, they necessarily share the same history. Phylogenetic analysis of the cyt b gene recovers the well-corroborated relationships, generally with strong support. None of the methods that we employed, including variously weighted parsimony, neighbor joining on both single-rate and gamma-corrected distances, and maximum likelihood, were able to recover these relationships for the 12S gene. Parsimony analyses of the 12S data resulted in a relatively strongly supported placement of Peromyscus eremicus that conflicts with that suggested by cyt b and all other data. There is extreme among-site rate variation in the 12S sequences and moderate levels in the cyt b sequences. This highly skewed distribution of rates in the 12S gene makes phylogenetic analyses of these sequences particularly susceptible to the misleading effects of nonindependence and other nonrandom noise, suggesting that phylogenetic analyses of data sets that contain a great deal of among-site rate variation be interpreted with caution.