We present a modelling system for the estimation of forest fire emissions (prebolchem-fire) and their inclusion in the atmospheric composition model BOLCHEM. Emission fluxes have been estimated following the methodology proposed by Seiler and Crutzen (1980) and using MODIS "Burned Area Product". Then they are modulated following the WRAP approach (WRAP Fire emission inventory 2005). This approach, integrated with information from a lagrangian backward model, is also used for the estimations of fire emission height. Model simulations have been performed for the period 22 - 30 August 2007, in which fires were most severe in Greece, Albania and Algeria. The emissions fluxes estimated for the gas species CO, NOx, SO2 and NH3 and the particulate matter PM2.5 and PM10 have been compared with those estimated by the global model FINNv1 (Fire Inventory from NCAR) and the difference between the two emissions have been discussed. To put into evidence the sensitivity of the results to the modulation of emission fluxes and height, modelling simulations have been performed with different model settings. Moreover, the concentrations of black carbon aerosol predicted by the model are compared with measurements at Tito Scalo, Potenza (40.63oN, 15.80oE, 760 m asl), where an unusual layer of aerosol was detected by the PEARL station lidar on 30 August 2007. We discuss the discrepancies between the measured and modelled vertical distribution concentrations of black carbon aerosol, probably due to uncertainties related to the estimation of the total flux and of the injection height of the smoke.