Abstract: Since proximal support vector machine(PSVM) is susceptible to uneven class sizes and is sensitive to outliers and noises in the training set, a robust PSVM was proposed. By imposing fuzzy memberships to each data point and introducing different error penalties for different classes, the robustness of PSVM was greatly enhanced. Both the affinity among samples and the relation between a sample and its class center were considered when calculating fuzzy memberships. Moreover, the similarity between the algorithm and ridge regression model was well demonstrated. Experiment results show that the robust PSVM has demonstrated enhanced classification ability.