The impact of surface ozone, PM2.5 and NO2 data assimilation in the current operational Canadian air quality model (GEM-MACH) is presented. A direct inversion algorithm has been used to ingest or assimilate observations and has been shown to produce satisfactory performance at low cost. Canadian observations are combined with US observations obtained from AIRNOW/USEPA data base. The positive impact of initializing the air quality model with objective analysis of surface pollutants on air quality forecast is fully evaluated and results are analyzed. Finally, inspection of the mean OmF (Observations minus Forecast) in terms of meteorological conditions indicate that the model bias varies significantly according to the different meteorological classes. The latter information can be used to develop effective algorithms for bias correction of pollutants for surface data assimilation and to evaluate model's weaknesses.