Abstract: Size loading is one of important indexes in evaluating the sizing quality,which is influenced by several process parameters,including four controllable factors,i.e.,sizing concentration,temperature in sizing box,speed of sizing machine and squeezing roller pressure.To establish a math model of the correlation of above four factors and the size loading for accurate prediction of size loading,a prediction model of three-layer Back-Propagation neural network system was established and the network was trained using the experienced data obtained from the practical production of pure cotton twill and repeated training was carried out with Levenberg Marquardt algorithms so that the performance goal was attained.This network model was applied to predict the size loading and the predicted results showed in good agreement with the actual value,and it can meet the requirements for production.