We describe the implementation and performances of a weather hindcast obtained by dynamically downscaling the ERA5 data across the period 1979-2019. The limited-area models used to perform the hindcast are BOLAM (with a grid spacing of 7 km over the Mediterranean domain) and MOLOCH (with a grid spacing of 2.5 km over Italy). BOLAM is used to provide initial and boundary conditions to the inner grid of the MOLOCH model, which is set in a convection-permitting configuration. The performances of such limited-area, high-resolution and long-term hindcast are evaluated comparing modelled precipitation data against two high-resolution gridded observational datasets. Any potential added-value of the BOLAM/MOLOCH hindcast is assessed with respect to ERA5-Land data, which are used as benchmark. Results demonstrate that the MOLOCH hindcast provides a lower bias than ERA5-Land as regards both the mean annual rainfall (-1.3% vs +8.7%) and the 90th percentile of summer daily precipitation, although a wet bias is found in southern Italy (bias \(\simeq\) +17.1%). Improvements are also gained in the simulation of the 90th percentile of hourly precipitations both in winter and, to a minor extent, in summer. The diurnal cycle of summer precipitations is found to be better reconstructed in the Alps than in the hilly areas of southern Italy. We also analyse rainfall peaks obtained in the simulation of two well-known severe precipitation events that caused floods and damages in north-western Italy in 1994 and 2011. We finally discuss how the demonstrated reliability of the BOLAM and MOLOCH models associated to the relatively low computational cost, promote their use as a valuable tool for downscaling not only reanalyses but also climate projections.