The problem of air pollution around urbanized area across Europe is strongly related to ozone. Ozone is a result of photo-chemical oxidation, and therefore an indication of the presence of pollutants. Overexposure to ozone is harmful to the health, and the concentrations are therefore measured on a regular basis to check exceedance of air-quality guidelines. Models have been developed to simulate the ozone formation, for example to make a forecast of the air quality for the coming days. A new direction in air-pollution modelling is data assimilation: merging model simulations and measurements in a single procedure. The target of a data assimilation problem is to decrease the difference between models and measurements, and with this, improvement of the simulations for which measurements are not available. This thesis describes the development of a data assimilation tool for an air pollution model, based on a Kalman filter. Given stochastic descriptions for the uncertainties in the model and the representation of the measurements, a Kalman filter is able to compute an optimal estimate of the pollutant concentrations. Experiments with the developed filter showed that the main uncertainties in the model are related to the resolution of parameters such as emissions and deposition, which is to coarse both in space as well as in time. The developed filter is able to compensate for these uncertainties by assimilation of ozone measurements. With this, the assimilation decreases the uncertainties in the model parameters, such that the filter could be used to estimate, for example, the actual emissions.