XRATE is a program for prototyping phylogenetic hiden Markov models and stochastic context-free grammars. It is used to discover patterns of evolutionary conservation in sequence alignments. The program can be used to estimate parameters for such models from "training" alignment data, or to apply the parameterized model so as to annotate new alignments. The program allows specification of a variety of models of DNA sequence evolution which may be arbitrarily organized using formal grammars.