We examine how well the National Center for Atmospheric Research (NCAR) regional climate model (RegCM2) simulates the mean and interannual variability of precipitation in a semiarid region to more fully establish the strengths and weaknesses of the model as a tool for studying regional scale climate processes. We compare precipitation observations with RegCM2 output from a 5.5 year long simulation of the climate of central Asia, driven by the European Center for Medium-Range Weather Forecasts analyses. RegCM2 simulates well the spatial patterns and annual cycles of precipitation observed in climatically different subregions. The magnitude of simulated precipitation is similar to observations except over the driest part of Central Asia where the simulated precipitation is too high. We calculate precipitation anomalies for each month as the difference between the monthly total and the 5 year average for that month, from both observations and RegCM2 output. The magnitude of simulated interannual variability is similar to observations, although there are differences. RegCM2 tends to underpredict (overpredict) the magnitude of variability in the same combinations of subregion and season for which it underpredicts (overpredicts) mean precipitation. RegCM2 closely reproduces precipitation anomalies observed in specific months, except during summer and during winter in the mountains. There is no correlation between model biases in mean precipitation and how well the model reproduces a series of precipitation anomalies. This suggests that the processes controlling the mean and the variability of precipitation differ. Therefore evaluating the ability of a regional climate model to simulate both quantities is a demanding test of model performance. ¿ 1999 American Geophysical Union