Two sets of experiments using a single-screw extruder were conducted with an ingredient blend containing 40% DDGS (distillers dried grains with solubles), along with soy flour, corn flour, fish meal, vitamin mix, and mineral mix, with the net protein content adjusted to 28%. The variables controlled in the first experiment included seven levels of die size, three levels of moisture content, three levels of temperature gradient in the barrel, and one screw speed. The variables altered in the second experiment included three levels of moisture content, three levels of temperature gradient in the barrel, five levels of screw speed, and one die size. Regression models and neural network (NN) models were then developed using the data pooled from the two experiments to predict extrudate properties and extrusion processing parameters. In general, both regression and NN models predicted the extrusion processing parameters with better accuracy than the extrudate properties. Similarly, lower R 2 values for the regression results corresponded to lower R 2 values in the NN modeling. The regression models predicted the extrusion processing parameters using three and six input variables with R 2 values of 0.56 to 0.97 and 0.75 to 0.97, respectively. The NN models predicted the extrusion processing parameters using three, five, and six input variables with R 2 values (between measured and predicted values) of 0.819 to 0.984, 0.860 to 0.988, and 0.901 to 0.991, respectively. With the regression modeling, even though increasing the number of input variables from three to six resulted in better R 2 values, there was no decrease in the coefficient of variation (CV) between the measured and predicted variables. On the other hand, the NN models developed with six input variables resulted in more accurate predictions with reduced CV and standard error. Because of its ability to produce accurate result with reduced variation and standard error, NN modeling has greater potential for developing robust models for extrusion processing.