A new descriptor of amino acids,SVREW, was derived from principal components analysis of 41 randic molecular profiles descriptors，44 eigenvalue-based indices descriptors and the matrix of 47 walk and path counts descriptors of amino acids.The structures of angiotensin-converting enzyme（ACE） inhibiting dipeptides and tripeptides，bitter tasting thresholds dipeptides and tetrapeptides，oxytocin，HLA-A*0201 restricted CTL epitope were characterized with SVREW，using multiple linear regression（MLR） to establish a quantitative structure-activity relationship，at the same time，adopting the methods of internal and external for dual verification of the stability of the model.The relevant statistical parameters were as follows：the correlation coefficient（R2cum），leava one out（LOO） cross-validation correlation coefficient（R2cv） and external validation correlation coefficient（Q2ext） were 0.994，0.797，0.948 for ACE inhibition dipeptides model；0.896,0.686,0.720 for ACE inhibition tripeptides models,0.955,0.859,0.879 for BTT dipeptides model,0.958,0.796,0.843 for BTT tetrapeptides model,0.990,0.954,0.890 for oxytocin model,0.950，0.748，0.773 for HLA-A* 0201 restricted CTL epitope model,respectively.Studies showed that the MLR models constructed by SVREW descriptor had good fitting and predictive abilities,which was an effective structure characterization method in peptide drugs QSAR study and provided a guidance for new drug discovery and research.