Turtle phylogeny remains uncertain despite a recent effort that analyzed a large data set combining mitochondrial DNA (mtDNA) sequence data from two genes and a suite of morphological and fossil characters (Shaffer et al. 1997). Existing uncertainties could represent the actual history of turtle evolution, or they may be inherent to the data themselves. To address this critical issue, we sequenced nearly all of the nuclear recombination activase gene 1 (RAG-1). RAG-1 provided a very "clean" data set in comparison to cytochrome b (the source of the majority of characters in Shaffer et al. (1997)) in terms of saturation, base composition bias, and base composition stationarity. Most of the relationships suggested by mtDNA/morphology analysis are also supported by RAG-1, though with a much higher level of support from RAG-1. RAG-1 and mtDNA/morphology disagree strongly in the placement of Chelydra and Platysternon, RAG-1 highly supports separation of the two taxa, while mtDNA/morphology strongly supports their sister relationship. Key morphological characters also support the separation of these two taxa, as do other sources of information, such as chromosomal features and biochemical properties. A combined analysis of the available nuclear DNA, mtDNA, and morphological characters (excluding Chelydra and Platysternon) provides a very robust evolutionary tree for turtles, with only one branch receiving <95% bootstrap proportion support. Using RAG-1 to estimate dates of divergence within the tree has a high degree of error, but is largely concordant with the sequence of appearance of taxa in the fossil record and indicates rapid radiations at the end of the Jurassic and in the middle of the Cretaceous. Almost all extant turtle families appear to have diverged prior to the Cretaceous-Tertiary boundary. RAG-1 exhibits a high degree of utility in resolving deep phylogenetic relationships among turtle families, as it has with other taxa.