Abstract: Deoxidization alloying is the last step of converter smelting process. The accuracy of alloy content control in molten steel directly affects the smelting difficulty and smelting cycle of refining process. The yield of alloy is an important reference standard for converter alloying workers. The accuracy of determining the yield of alloying elements directly affects the stability of molten steel composition and production cost. Through theoretical analysis and actual data verification，9 observable indicators that affecting manganese yield were selected. The data was then reduced by means of the factor analysis，and the 6 public factor matrices were obtained as the input of the model，the manganese yield was the output of the model，and the manganese 
yield prediction model based on GA-BP neural network was established. The results show that the regression coefficient of the model R² = 0.714 78，the average error is 0.01. The number of more than 98 % of the predicted accuracy accounts for 75 % of the total，and the accuracy of the prediction is high，and the actual production has certain reference significance. 

Key words: manganese yield, factor analysis, predictive model, influencing factors, BP neural network