Constitutive relationship of metallic rubber was nonlinear. Considering the constitutive relationship’s complication and BP neural net’s good ability to dispose of nonlinearity, it was necessary that constitutive relationship of metallic rubber on basis of BP neural net was studied. In this paper, coefficients of constitutive relationship for metallic rubber were studied and trained by BP neural net for the two conditions, in which shape factor is only various, density and shape factor are both various, and then coefficients of constitutive relationship were obtained. Coefficients from net prediction were compared with coefficients from experimental data fitted, and they had better consistence. It was proved that prediction for constitutive relationship of metallic rubber by BP neural net was reasonable for the two conditions, in which shape factor is only various, density and shape factor are both various.