Abstract: This paper aimed to study the effects of different replacement ratios of cottonseed concentrated protein replacing fish meal and processing parameters on the quality of low-starch slow-sinking extruded feed, and to optimise and predict the suitable processing parameters and pellet quality with different replacement ratios. A Box-Behnken design was adopted, with the replacement ratio (0, 25%, 50%), die temperature (100, 120, 140 °C), and the moisture content of mash feed after conditioning (25%, 27%, 29%) as independent variables in this research. The bulk density, sinking speed, sinking ratio at 10 min and specific mechanical energy (SME) of slow-sinking extruded feed as dependent variables in this study. A total of 17 extruded processing experiments were conducted. The correlation model between processing parameters and the quality of low-starch slow-sinking extruded feed, as well as the prediction model of product quality were established using response surface methodology and regression analysis. The results showed that the bulk density, sinking speed and sinking ratio at 10 min decreased gradually with the increase of the replacement ratio, while SME increased gradually. With the increase of the die temperature, the bulk density, sinking speed and sinking ratio at 10 min first increased and then decreased, while SME first decreased and then increased gradually. With the increase of moisture content of mash feed after conditioning, the bulk density and sinking ratio at 10 min increased gradually. In contrast, the sinking speed first decreased and then increased, and SME decreased gradually. The results of variance analysis indicated that the order of effect of each factor on the quality of low-starch slow-sinking extruded pellet feed was as follows: the replacement ratio of cottonseed concentrated protein > the moisture content of mash feed after conditioning > the die temperature. The suitable processing parameters and pellet quality with different replacement ratios of cottonseed concentrated protein replacing fish meal were further optimised and predicted. With the increase of the replacement ratio, the operable range first increased and then decreased. When the replacement ratio exceeded 15%, the processing difficulty of the material gradually increased. The optimisation results of the suitable processing parameters and prediction model of product quality for the material with a cottonseed concentrated protein replacement ratio of 29% were verified. The relative errors between the predicted values and the experimental values were all less than 5%, indicating that the model was reliable and the optimised processing parameters were accurate and reliable. In conclusion, using cottonseed concentrated protein to replace fish meal (replacement ratio <15%) and appropriately increasing the moisture content of the conditioning mash feed, as well as reducing the die temperature, is conducive to the processing and production of low-starch slow-sinking extruded pellet feed.