This study evaluates the capabilities of the Weather Research and Forecasting (WRF) model to reproduce the space-time variability of near-surface air temperature over Tunisia. Downscaling is based on two nested domains with a first domain covering the Mediterranean Basin and forced by 21 years of ERA-Interim reanalysis (1991-2011), and a second domain (12 km spatial resolution) centered on Tunisia. Analyses and comparisons are focused on daily average (Tavg), minimum (Tmin) and maximum (Tmax) near-surface air temperatures and are carried out at the annual and seasonal timescales. WRF results are assessed against various climatological products (ERA-Interim, EOBS and a local network of 18 surface weather stations). The model correctly reproduces the spatial patterns of temperature being significantly superimposed with local topographic features. However, it broadly tends to underestimate temperatures especially in winter. Temporal variability of temperature is also properly reproduced by the model although systematic cold biases mostly concerning Tmax, reproduced throughout the whole simulation period, and prevailing during the winter months. Comparisons also suggest that the WRF errors are not rooted in the driving model but could be probably linked to deficiencies in the model parameterizations of diurnal/nocturnal physical processes that largely impact Tmax / Tmin.