[Objectives] In view of the fact that the backfat thickness of pig carcass is still measured manually in most slaughtering enterprises, and then pig carcasses are graded manually according to their weight and the backfat thickness, which results in such problems as labor intensive, time-consuming and high risk of disease transmission between humans and animals. This paper aimed to establish a prediction model for pig carcass weight. This model could be applied to predict the pig carcass weight and automatically grade pig carcass by using image processing and other technologies to obtain relevant parameters of the model from carcass. [Methods] Twenty pig carcasses removed hooves with weight 50-90 kg were selected respectively in the three periods of 14:00-15:00, 15:20-16:20 and 16:30-17:30 after pig had been slaughtered within 15 min according to standardized technology. The models for pig carcass weight prediction were established, optimized, and accuracy verified, on the basis of the measured parameters of carcasses' three widths (Lf, L1/2 and Lr), length (Lt), backfat thickness (t1/2)and weight (w). [Results] Test results showed that, by using weighted mean value of carcasses (Le) instead of t1/2, the carcass weight prediction model of w=4.05Le+0.45Lt-116.32 consisted of Lt, Le and constant, its correlation coefficient (R²)increased from 0.48 to 0.96, and significant coefficient was 0.01, and the prediction accuracy reached 94.16%. [Conclusions] The accuracy of the carcass weight prediction model established was further improved compared with other models due to reducing the error of the width by using weighted mean value. [ABSTRACT FROM AUTHOR]