In this study, multiple machine learning approaches, including support vector machine (SVM), k-nearest neighbor (k-NN), artificial neural networks (ANN) and logistic regression (LR), are applied for classification of HIV-1 protease inhibitors(PIs) from molecular structure. A diverse set of 641 compounds, including 414 active agents (PIs þ ) and 227 inactive agents (PIs À ), are adopted to develop the classification models. A hybrid feature selection method, which combines Fischers score and Monte Carlo simulated annealing embedded in the support vector machine approach, is used to select the relevant descriptors from a pool of 1559 molecular descriptors. Three validation methods are employed to validate the model in this study. The first one is the five-fold cross validation method and the averaged prediction accuracies for these machine learning approaches are between 83.9 -93.5% for PIsþ and between 67.0 -77.7% for PIsÀ agents. The second validation method is the external test set and the prediction accuracies for PIsþ are between 84.6 -95.2% and for PIsÀ agents are between 63.2 -87.7%. These two validation methods show that the SVM model has better overall performance than other three machine learning models. The third validation method is the yscrambling method, which shows no obvious chance correction in the developed SVM model. The prediction method proposed in this work can give better generalization ability than other recently published methods and can be used as an alternative fast filter in the virtual screening of large chemical database.