Where air pollution control is concerned, the rare (extreme) event is typically more significant than the common event. Conventionally, a theory developed to address questions relating to the distribution of extremes is the Extreme Value Theory (EVT). The Bayesian approach offers a more comprehensible framework in incorporating all of the uncertainties involved in the prediction process using the conventional methods. To evaluate the performances of the classical and Bayesian approaches using non-informative priors in estimating the parameters of the Generalized Extreme Value (GEV) and to attain the best model to predict PM10concentrations level. The daily maximum monitoring records of PM10from January 2000 to December 2012 furnished by the Department of Environment, Malaysia were used in this study. The goodness-of-fit of the distribution was determined using performance indicators, namely, the accuracy measures and error measures. The best distribution was selected based on the highest accuracy measures and the smallest error measures. This study revealed that the Bayesian GEV with non-informativeprior gave the best estimators for PM10concentrations in three industrial monitoring stations and it could be applied in the PM10analysis to predict the exceedances of future extreme concentrations of PM10in these stations.