A Gram-negative bacterial protein can be resident at one of five primary subcellu-lar localizations. Proteins resident at outer membrane are called outer membrane pro-teins (OMPs). Because of their exposition at the surface of the bacterial cell, these pro-teins attract the research interest of drug target. Since identifying OMPs by experiments takes lengthy time, it is urgent to develop reliable methods to discriminating OMPs from other proteins. In this thesis, we present a method for OMP prediction by Support Vector Ma-chines (SVMs) using the combinations of k-gapped amino acid pair compositions. Two dataset are used to evaluate our method. One dataset consists of 471 OMP sequences and 1,120 non-OMP sequences. These sequences are annotated with experimentally verified subcellular localizations. The other dataset consists of 377 OMPs and 1,120 globular proteins belonging to four typical structural classes. Using the former dataset, our classifier achieves 95% in precision and 92% in recall. The result indicates that the combination of k-gapped amino acid pair compositions captures more discriminatory information than the occurrences of frequent subsequences. Applied to the latter dataset, our classifier achieves as high as 96% in precision and recall. Compared to the statisti-cal method based on dipeptide composition, the result indicates that our SVM based method is better than the pure statistical method. Furthermore, our classifier performs well on an extended dataset containing additionally α-helical transmembrane proteins.