One of the major drawbacks of conventional air quality models is their inability in accurately predicting extreme air pollutant concentrations. Hybrid modelling is one of the techniques that estimates/predicts the ‘entire range’ of the distribution of pollutant concentrations by combining the deterministic based models (capable in predicting average range) with suitable statistical (probability) distribution models (capable in predicting extreme range). This research paper describes system based approach in developing hybrid model to predict hourly averages as well as extreme percentile ranges of NOx and PM2.5 concentrations at two urban locations having complex traffic heterogeneity, highly variable tropical meteorology and different geographical characteristics. At one of the selected locations i.e. Delhi megacity, during winters, hybridization of AERMOD and Lognormal predicts NOx and PM2.5 concentrations satisfactorily with index of agreement ‘d’ values of 0.98–0.99, respectively; however, during summers, AERMOD-Log-logistic and AERMOD-Lognormal are best predicting NOx and PM2.5 concentrations with d values of 0.98–0.96, respectively. In another, i.e., Chennai, a coastal megacity, AERMOD-Lognormal predicts PM2.5 concentrations satisfactorily with d values of 0.98 and 0.99 during winter and summer seasons, respectively. Further, hybrid model has also been used to evaluate regulatory compliance. © 2017 Elsevier Ltd