Regional Climate Models (RCMs) are the primary climate information available to public stakeholders and city-planners to support local adaptation policies. However, with resolution in the order of ten kilometres, RCMs do not explicitly represent cities and their influence on local climate (e.g. Urban Heat Island; UHI). Downscaling methods are required to bridge the gap between RCMs and city scale. A statistical–dynamical downscaling methodology is developed to quantify the UHI of the city of Paris (France), based on a Local Weather Types (LWTs) classification combined with short-term high-resolution (1-km) urban climate simulations. The daily near-surface temperature amplitude, specific humidity, precipitation, wind speed and direction simulated by the RCMs are used for the LWTs attribution. The LWTs time series is associated to randomly selected days simulated with the mesoscale atmospheric model Meso-NH coupled to the urban canopy model Town Energy Balance to calculate the UHI corresponding to the successive LWTs. The downscaling methodology is applied to the EURO-CORDEX ensemble driven by the ERA-Interim reanalysis, and evaluated for the 2000–2008 period against station observations and a 2.5-km reanalysis. The short-term dynamical simulations slightly underestimate and overestimate near-surface minimum and maximum air temperature respectively, but capture the UHI intensity with biases in the order of a tenth of a degree. RCMs show significant differences in the variables used for the LWTs attribution, but the seasonal LWT frequencies are captured. Consequently, the reconstructed temperature fields maintain the small biases of the Meso-NH simulations and the statistical–dynamical downscaling greatly improves the UHI compared to the raw data of RCMs.