Traditionally drug discovery has been a labor intensive effort, since it is difficult to identify a possible drug candidate from an extremely large small molecule library for any given target. Most of the small molecules fail to show any activity against the target because of electrochemical, structural and other incompatibilities. Virtual screening is an in-silico approach to identify drug candidates which are unlikely to show any activity against a given target, thus reducing an enormous amount of experimentation which is most likely to end up as failures. Important approaches in virtual screening have been through docking studies and using classification techniques. Support vector machines based classifiers, based on the principles of statistical learning theory have found several applications in virtual screening. In this paper, first the theory and main principles of SVM are briefly outlined. Thereafter a few successful applications of SVM in virtual screening have been discussed. It further underlines the pitfalls of the existing approaches and highlights the area which needs further contribution to improve the state of the art for application of SVM in virtual screening.