A new amino acid structure descriptor,namely SVHEHS,was derived from the principal component analysis of 457 physicochemical properties of natural amino acids.The descriptor was used to characterize the structures of angiotensin converting enzyme(ACE） inhibitory dipeptides,tripeptides and tetrapeptides,and then the mathematic models between the structures and the activities of three panels of peptides were established by artificial neural networks.The correlative coefficient(r2）,the cross-validation correlative coefficient(Q2LOO）,root mean square error(RMSE）,external validation correlative coefficient(Q2ext） for the dipeptides model were 0.946,0.951,0.249 and 0.852,respectively,for tripeptides model were 0.973,0.945,0.135 and 0.813,respectively,and for tetrapeptides model were 0.915,0.879,0.250 and 0.814,respectively.The result indicated that the models based on SVHEHS descriptor combined with neural networks had good fitting and predictive abilities.Moreover,the key structure factors relevant with peptide activities were studied by the mean impact value method.The results can be helpful for the molecular designs of new ACE inhibitory peptides.