By incorporating class-invariant prior information into the object function of a Support Vector Machine (SVM) algorithm, a SVM infrared spectroscopic qualitative analysis algorithm based on drift constraint is proposed. Because the algorithm simulates the drift term of infrared spectrum into a low order polynomial and requires the normal vector of the decision surface to be perpendicular to the drift direction, the classifier can remove the effect of sample drift. The influence of band selection and regularization parameters on classification accuracy is described in detail and the classification results of different SVM algorithms are compared. The experimental result shows that compared with the standard SVM algorithm and other similar algorithms, the DCSVM algorithm has a higher classification accuracy.