Discriminating cell wall lytic enzymes from non lytic enzymes is a very important task for curing bacterial infections. In this paper, based on Chou's amphiphilic pseudo amino acid composition, we develop fisher-discriminant based classifier to predict cell wall lytic enzymes. Experiments show that 66.7% sensitivity with 88.6% specificity is obtained. The method is further able to predict endolysin and autolysin with an overall accuracy of 92.9%. Results demonstrated that our method can provide highly useful information for further bacterial control research.