The impact of 3D-Var on 2.5 km convective-scale forecasts is investigated by using the WRF and HARMONIE models. The numerical simulations were carried out for the period of 1 to 15 Dec 2013. The domain of simulations is a small domain in the west of Iran. The LBCs were exploited from the ECMWF operational forecasts. Furthermore, conventional observations along with tuned domain specific background error statistics (BES) were used. Generally, the superiority of the HARMONIE was seen in the sense that it produces more precise forecasts than the WRF over the selected area during the chosen period, even without using observations in the data assimilation (DA). Results show the advantage of using DA for temperature forecasts. This positive effect was more pronounced for the HARMONIE compared to the WRF forecast, probably due to using soil DA. Furthermore, a near neutral effect was seen for surface pressure and relative humidity.