The polynomial regression model is a widely applied nonlinear regression method. Since the high correlation exists among independent variables in the polynomial regression model, it will induce excessive computational error to estimate coefficients with the ordinary least square regression. A method of polynomial regression modeling based on Gram-Schmidt process which can achieve the orthogonalization of the independent variables and overcome the adverse effects of multicollinearity to regression modeling was proposed, so as to apply ordinary least square to regression modeling effectively. The independent variables including notable explaining information can be selected effectively, at the same time redundant information is deleted. Simulation data analysis was adopted to test the effectiveness of the method.