An air quality management system has been in operational use in the city of Stockholm, Sweden, for several years. In the system there are, in addition to databases for measurements of air pollution and meteorological parameters and emission data, several dispersion models. The models include a Gaussian and a Eulerian (grid) type of dispersion model. An evaluation of the Gaussian and the grid models for Stockholm has been carried out. Simulated concentrations of nitrogen oxides, NOx, have been compared with measured NOx concentrations at three measuring stations in central parts of the city. The evaluation was based on comparisons between measured and simulated concentrations in two time-series, a summer week and a winter week, and between two seasonal simulations, a summer and a winter seasonal average and extreme concentrations. It was noted that the mean difference between the measured and the simulated concentrations during all trials varied from a few percent up to 60%. It was also noted that the differences for the seasonal averages and 98-percentile decreased considerably when the simulated value was picked from the gridpoint with best agreement with the measured value, inside a circle with a radius of one grid distance from the measuring station.