XRATE implements algorithms for comparative annotation, ancestral reconstruction, evolutionary rate estimation, and simulation. Its modeling repertoire includes phylogenetic stochastic context-free grammars and phylo-hidden Markov models. Following earlier tests of XRATE as a machine-learning tool suitable for alignment annotation, we now report the first tests of XRATE as a precise quantitative instrument for estimating evolutionary rates. We implement a codon model similar to that of Goldman and Yang (1994) (A codon-based model of nucleotide substitution for protein-coding DNA sequences. Mol Biol Evol 11: 725-736) and show that XRATE's parameter estimates are consistent with those of PAML. To demonstrate its utility, we apply the model to measure the difference in selective strength (omega) between intracellular and secreted regions of type I transmembrane proteins. In 215 of 303 instances, a complex model with individual omega for each region provides a better fit to the data than the simpler single omega value model. Secreted portions of type I transmembrane proteins show an elevation in omega similar to that seen for secreted protein genes. Less stringent purifying selection is thus a general property of the extracellular milieu, rather than being specific to only soluble and secreted proteins.