Reproduction of daily maximum and minimum temperatures, including tails of their distributions and links to large-scale circulation, is evaluated in an ensemble of high-resolution regional climate model (RCM) simulations over the Czech Republic. RCM data for recent climate (1961–1990) are validated against observed data gridded from a high-density station network. We find large biases in mean monthly temperatures and in seasonal extremes, which are significant in most RCMs throughout the year. The results suggest that an RCM's formulation plays a much more important role in summer, whereas in winter RCM performance is closely linked to the driving GCM. Biases are usually larger for extremes than central parts of temperature distributions, and RCMs tend to underestimate the severity of extremes in both seasons. Substantial underestimation of diurnal temperature range throughout the year in all RCMs and a shift of maximum in its annual cycle suggest general errors in simulating climate processes affecting the difference between daytime and nighttime temperatures. Some features of the temperature biases in RCMs are related to deficiencies in the simulation of atmospheric circulation, particularly too strong advection and overestimation of westerly flow at the expense of easterly flow in most RCMs. The general biases in simulating anticyclonic, cyclonic and straight flow also contribute to the underestimated diurnal temperature range.