We present three di erent approaches to model extreme values of daily air pollution data. We tted a generalized extreme value distribution to the monthly maxima of daily concentration measures. For the exceedances of a high threshold depending on the data the parameters of the generalized Pareto distribution were estimated. Accounting for autocorrelation clusters of exceedances were used. To get information about the relationship of the exceedance of the air quality standard and possible predictors we applied logistic regression. Results and their interpretation are given for daily average concentrations of O 3 and of NO 2 at two monitoring sites within the city of Munich.