Gene-encoded antimicrobial peptides (AMPs) are widely distributing in all classes of life ranging from plants to animals, including amphibians, birds, fish, and mammals etc. The characteristics of wide-spectrum, rapidness and specificity, and the activities against several bacteria, fungi, viruses, protozoa and cancerous cells, allow the development of numerous antimicrobial peptides with potentially useful properties as therapeutic agents. Development of activity predicted tools based on understanding the role of relationship of structure-activity is inevitable for their drug designs. In this study, we manually collected 1162 antimicrobial sequences firstly, then builded a comprehensive design platform for studying structure-activity relationship of antimicrobial peptides, finally constructed a new activity prediction model with stepwise discriminant analysis. This model integrated seven physicochemical parameters, including length, molecular weight, theoretical pI, primary amino acid composition, Instability index, aliphatic index and grand average of hydropathicity, charge. The prediction model correctly classified 70% of the known activity peptides in the database, and predicted 77.8% of the new unknown activity peptides. The results indicated that the model consisting of four discriminant functions can recognize and classify activity for antimicrobial peptides effectively.