【Objective】The binary quadratic polynomial fertilizer response model is set up on the assumption that crop yield increment per unit of nutrient is linearly related to fertilizer application rate, thus leading to a symmetric relationship in crop response to fertilization between before the maximum application rate and after the maximum application rate of fertilizer is applied. This model obviously tends to be low in modeling success ratio mainly because of its set biased error and existence of multicollinearity and heteroscedasticity. This study has developed a binary non-structural fertilizer response model, which is designed to have better applicability. 【Method】On the basis of the unary non-structured fertilizer response model and the principle that the functions of plant nutrient elements cannot be replaced by each other, a binary non-structural fertilizer response model is constructed. Its fitting effect and its reliability in recommendation of fertilization rates is verified via NPK binary combination field experiments on crop response. 【Result】 The model is OKed through all the statistical significance tests in all the 17 field experiments on crop response of rice, peanut, potato, edamame, winter wheat and summer maize, but only 58.8% of the total test sites showed typical crop responses, and 41.2% did non-typical ones with irrational algebetic symbols of monomial or quadratic term coefficient in model and extrapolated fertilizer recommendation. Fitting tests of the model demonstrate that it has been OKed through all the statistical significance tests, with 88.2% of the test sites showing typical crop responses, and only 11.8% non-typical ones, all of the type of extrapolated recommendation, which demonstrates that the new model has improved in modeling success rate by 29.4% as compared with the binary quadratic polynomial fertilizer response model. Both models show significant linear positive relationships between maximum fertilization rate for highest crop yield and economic fertilization rate for highest crop yield, but the first term coefficient of the linear regression equation is 0.915 3 and 0.916 1, only, which means that for the increment of each kg of fertilizer recommended by the binary quadratic polynomial fertilizer response model only 0.915 3 kg and 0.916 1 kg is recommended by the new model. The new model has fairly well overcome the problem of the quadratic polynomial model’s recommendation tending to be higher. 【Conclusion】 All the analyses indicate that the binary quadratic polynomial fertilizer response model is a simple and specific one of the binary non-structural fertilizer response model. The new model has a wider range of applications.