To investigate the evolutionary dynamics of rabies virus (RABV) in China, we collected and sequenced 55 isolates sampled from 14 Chinese provinces over the last 40 years and performed a coalescent-based analysis of the G gene. This revealed that the RABV currently circulating in China is composed of three main groups. Bayesian coalescent analysis estimated the date of the most recent common ancestor for the current RABV Chinese strains to be 1412 (with a 95 % confidence interval of 1006-1736). The estimated mean substitution rate for the G gene sequences (3.961¾10"4 substitutions per site per year) was in accordance with previous reports for RABV.Rabies has become an important public health concern in countries in Asia and Africa. It is estimated that this disease is responsible for over 55 000 human deaths annually, of which approximately 56 % occur in Asia and 44 % in Africa (WHO, 2005). Rabies was first described in China in about 556 BC and has been continually reported for the following 2500 years (Wang & Huang 2001 et al., 2005). In this study, we used Bayesian MCMC methods to analyse sequences from the highly variable G gene of RABV to determine phylogenetic relationships. By using G gene data collected over several years from almost all regions that had experienced cases of rabies, we investigated the evolution of RABV variants and determined the population history of RABV circulating in China.A total of 55 complete G gene sequences (1572 nt) isolated from four hosts (dog, human, cattle and deer) from 14The GenBank/EMBL/DDBJ accession numbers for the 25 RABV G gene sequences isolated in this study are FJ602447-FJ602459 and FJ418876-FJ418887. provinces, administrative municipalities and autonomous regions of China were analysed. The four hosts from which the viruses were isolated represent reservoirs and spillover hosts. Of these isolates, 25 were newly sequenced as part of this study (GenBank accession nos FJ602447-FJ602459) by using methods described previously (Meng et al., 2007); the remaining 30 isolates were obtained from GenBank. The year of isolation was available for all the isolates; full details are provided in Fig. 1 and Table 1.Estimates of the rate of molecular evolution, m (substitutions per site per year), and the time to most recent common ancestor (TMRCA) for the complete G gene alignment were obtained using a Bayesian MCMC method which was implemented in the BEAST program (available at http://beast.bio.ed.ac.uk/Main_Page). The best-fit model of nucleotide substitution for Bayesian analysis was selected with the MODELTEST software using Akaike's Information Criterion (AIC) (Posada & Crandall, 1998). The general time reversible (GTR) substitution model, incorporating a proportion of invariable sites (I) and a gamma distribution of rate variation among sites (C4) was used for the BEAST analysis. We employed both strict and relaxed (uncorrelated exponential and lognormal) molecular clocks (Drummond et al., 2006) to explore the extent of variation in the rate of nucleotide substitution and ...