P-Glycoprotein (P-gp) contributes to extruding a structurally, chemically, and pharmacologically diverse range of substrates out of cells. This function may result in the failure of chemotherapy in cancer and influence pharmacokinetic properties of many drugs. Although a great deal of research has been devoted to the investigation of P-gp and its substrate specificity, still we do not have a clear understanding of the resolution of the three-dimensional structure of P-gp and its working role as a drug efflux pump at a molecular level. Hence to identify whether a compound is a P-gp substrate or not, computational methods are promising both in cancer treatment and the drug discovery processes. We have established more effective models for prediction of P-gp substrates with an average accuracy of >90% using a Particle Swarm (PS) algorithm and a Support Vector Machine (SVM) approach. The applied models yielded higher accuracies and contained fewer variables in comparison with previous studies. An analysis of P-gp substrate specificity based on the data set is also presented by a PS and a SVM. The aim of this study is 3-fold: (i) presentation of a modified PS algorithm that is applicable for selection of molecular descriptors in quantitative structure-activity relationship (QSAR) model construction, (ii) application of this modified PS algorithm as a wrapper to undertake feature selection in construction of a QSAR model to predict P-gp substrates with a multiple linear (ML) and SVM approach, and (iii) also finding factors (molecular descriptors) that most likely are associated with P-gp substrate specificity by using a PS and a SVM from the data set.