Abstract: Develop ANN learn-forecast system by employing BP algorithm to forecast the performance of IF steel, test and analyze the system by using data collected from BAO Steel, and compare the precision of forecasted data with that of the multivariant linear regression model. The results show that the relative errors of ANN learn-forecast system on σ_b,δ₁, r and n are all less than 5.0% except that on σ_0.2 is 9.0%. It is concluded that this system has a higher forecast precision than the multivariant linear regression model.