Two of the critical-to-quality characteristics (CTQs) in the end milling of carbon fiber reinforced polymer (CFRP) composites are delamination factor (FD) and surface roughness (SR). The target of these two CTQs is smaller is better. The minimum FD and SR can be achieved by selecting the correct end milling parameters such as the speed of spindle and feeding speed, and also depth of cut. In this study, the minimization of FD and SR is conducted by applying a combination of backpropagation neural network (BPNN) and genetic algorithm (GA). BPNN is utilized to develop the model of FD and SR, while GA has two purposes. The first purpose of GA is to determine the BPNN network architecture that can produce a minimal mean square error (MSE) to precisely predict FD and SR. The second purpose is to find the best setting of end milling parameters (spindle speed, feeding speed, and depth of cut) that can minimize FD and SR simultaneously during the end milling process of CFRP composite. This study uses a randomized full-factorial 2 × 3 × 3 experimental design. The depth of cut of end milling parameters has two levels, while spindle speed and feeding speed each has three levels. The best BPNN network structure consists of three neurons in input layers, three neurons in one hidden layer, and two neurons in output layer; with tangent sigmoidal as the activation function. The setting of the end milling parameters that can minimize FD and SR simultaneously in end milling process of CFRP composites by using BPNN-GA is spindle speed of 3425 rpm, feeding speed of 29.4 mm/min, and depth of cut 1 mm.