Mercury emission from coal combustion has become a global environmental problem. In order to accurately reveal the complexly nonlinear relationships between mercury emissions characteristics in flue gas and coal properties as well as operating conditions, an alternative model using support vector machine (SVM) based on dynamically optimized search technique with cross-validation, is proposed to simulate the mercury speciation (elemental, oxidized and particulate) and concentration in flue gases from coal combustion, then the configured SVM model is trained and tested by simulation results. According to predicted accuracy of indicating generalization capability, the model performance is compared and evaluated with the conventional multiple nonlinear regression (MNR) models and the artificial neural network (ANN) models. As a result, it is found that, the SVM provides better prediction performances with the mean squared error of 0.0095 and the correlation coefficient of 0.9164 for testing sample. Moreover, based on the SVM model, the correlativity between coal properties as well as operating condition and mercury chemical form is also analyzed in order to deeply understand mercury emissions characteristics. The result demonstrates that SVM can offer an alternative and powerful approach to model mercury speciation in coal combustion flue gases. (C) 2009 Elsevier B.V. All rights reserved.