Regional climate models (RCMs) provide critical detailed information about the local climate. However, running RCM simulations requires powerful computers and is computationally expensive. We present a probabilistic RCM emulator. The model is called MESMER-RCM. It is a data-driven statistical model, designed to replicate RCM simulations with minimal computational resources. MESMER-RCM can generate large ensembles of synthetic, yet physically plausible fine-scale 2-meter temperature projections spanning multiple decades. These projections are very similar to those of real RCMs and effectively capture the internal variability implied by the RCM.