We investigated the use of support vector machines (SVMs) to distinguish quasars from stars using survey databases from different wavebands. We first employed a random forest approach for selection and weighting of features. We find that SVMs are an effective method to preselect quasar candidates from multiwavelength data.