QSAR models of 22 benzamidine derivatives reported as inhibitors of thrombin have been developed using the descriptors heat of formation, valence connectivity index, shape index, solvent accessibility surface area, molar refractivity, log P and molecular weight. QSAR models, in which either heat of formation or shape index or molar refractivity is present, have good predictive powers as correlation coefficients have been found above 0.9. These descriptors alone provide good QSAR models. Best QSAR model with 0.994001 value of correlation coefficient has been obtained using the combination of descriptors heat of formation, shape index, molar refractivity and log P.