Membrane proteins, which occupies the proportion between 20 and 35% of whole genomes, play various roles indispensable for organisms such as receptor and channels. Since experimental structural analysis of membrane proteins is experimentally very difficult, computational tools are strongly required for extracting the information about the structure of membrane proteins from amino acid sequences. Previously, we developed a membrane protein prediction system SOSUI [1], which provides the information about the existence, the number and the sequence regions of transmembrane helices. However, this system cannot predict the topology of transmembrane segments nor the existence of signal peptides. A signal peptide is located at the amino terminal of an amino acid sequence of a secretion protein or a secretion protein. Signal peptides are usually as hydrophobic as most transmembrane helices. Their main function is to transmit the carboxyl side of an amino acid sequence to the external media of a cell. Signal peptides are cut off from the rest polypeptide after the transfer of the carboxyl part. However, some of hydrophobic segments, which are called signal anchors, are not cut off. Therefore, for the prediction of signal peptides, we have to predict the signal sequences, which are hydrophobic enough to be penetrated into membrane, and also to discriminate signal peptides from signal anchors. In this work, we develop a system, which combines a novel system for the signal peptide prediction with the previous system SOSUI. Two kinds of amino acid indices were developed for the signal sequence prediction and the discrimination between signal peptides and signal anchors. The input data of the system is amino acid sequences alone and the output is the existence of a signal peptide, the discrimination of membrane proteins and the prediction of the region of a signal peptide as well as transmembrane helices. The accuracy of signal peptide prediction was better than 80%.