In this study, an attempt has been made to predict the major functions of gramnegative bacterial proteins from their amino acid sequences. The dataset used for training and testing consists of 670 non-redundant gram-negative bacterial proteins (255 ofcellular process, 60 of information molecules, 285 of metabolism, and 70 of virulence factors). First we developed an SVM-based method using amino acid and dipeptide composition and achieved the overall accuracy of 52.39% and 47.01%, respectively. We introduced a new concept for the classification of proteins based on tetrapeptides, in which we identified the unique tetrapeptides significantly found in a class of proteins. These tetrapeptides were used as the input feature for predicting the function of a protein and achieved the overall accuracy of 68.66%. We also developed a hybrid method in which the tetrapeptide information was used with amino acid composition and achieved the overall accuracy of 70.75%. A five-fold cross validation was used to evaluate the performance of these methods. The web server VICMpred has been developed for predicting the function of gram-negative bacterial proteins (http://www.imtech.res.in/raghava/vicmpred/).