In this paper, we propose to extend the multisurface proximal support vector machines to the problem of outlier detection. Instead of considering two non parallel proximal planes for extracting classes, we only seek a plane which is proximal to the target or dominant population and as far as possible from outliers. From this result, we show that a simple modification of the criterion introduces an effective contrast measure to isolate a target or dominant data population from outliers. Introducing the kernel trick, we extend the proposed algorithm to nonlinear data sets. The proposed algorithm is compared with recent novelty detectors on synthetic and real data sets.