In this study, an adaptive neuro-fuzzy logic method has been proposed to estimate roadside NO2 concentration levels. In the analysis, data from summer and winter seasons were modeled separately and five statistical measures, namely, RMSE, IA, R-2, NMSE and FB, were used for modeling evaluation. The available data (N=5797) for 2003 were divided into three categories: training, testing and checking, to set up the ANFIS model. The model was trained using 4923 data with 13 input variables consisting of air quality and meteorological data. Summer season data set (between July and August, N=361) and winter season data set (between December and February, N=361) have been separately used for prediction (testing) purposes. In general, RMSE (4.78 and 4.53), NMSE (0.029 and 0.026) and FB (0.03 and 0.01) values are low but IA (0.96 and 0.98) and R-2 (0.92 and 0.95) are reasonably high enough to predict the observed values for winter and summer season test data, respectively. In addition, the FOEX values show that the model slightly under-predicts for all input parameters. Overall, the statistical measures confirm the adequacy of the model for predicting NO2 levels in M25 Roadside for winter and summer season test data.