We analyze sensitivity experiments in which random perturbations are applied to the initial conditions (IC) and lateral boundary conditions (LBC) of a set of seasonal regional climate model (RCM) simulations. The model response to these perturbations, which is a measure of the internal RCM variability, grows for the first 5--15 days of simulation and then reaches a dynamical equilibrium. The response is not sensitive to either the magnitude or the source of the perturbation, but it is mostly tied to synoptic conditions, season (maximum response in summer and minimum in winter), region, and model configuration. The model response does not affect significantly the domain-wide average climatology, but it substantially influences the day-to-day model solution, especially precipitation, and aspects of the model climate such as the frequency of occurrence of heavy precipitation events. Results from the perturbation experiments are compared to those from an experiment in which some land surface conditions are modified. It is shown how some of the model responses to the physical forcing are within and some outside the internal model variability. Our results indicate that the internal model variability is a factor that should be considered in the design, analysis, and interpretation of RCM experiments. ¿ 2000 American Geophysical Union