The most problematic defect during CFRP drilling process was delamination in the hole entry and exit since delamination can lead to material failure and reduce the lifespan of component. The hole entry delamination (FDentry) and hole exit delamination (FDexit) are the measured response to appraise the quality of the drilling process. The critical quality to the characteristics of these responses is "smaller-is-better." The objective of the present study is to attain the best combination of drilling parameters in order to obtain the minimum FDentry and FDexit during CFRP drilling process. Three drilling parameters, namely drill bit geometry (Pa), spindle speed (n), and feeding speed (Vf), were utilized as the input parameters. Both spindle speed and feeding speed have three levels, and drill bit geometry has two levels. In the present study, a full factorial design experiment with three replications was used. The optimization of delamination in the hole entry and exit were conducted by utilizing the integration of a backpropagation neural network method and particle swam optimization (BPNN and PSO). The finest BPNN network architecture that can model the CFRP drilling process and predict the delamination in the hole entry and exit is one hidden layer with twelve neurons, three neurons on input layer and two neurons on output layer or 3-12-2. The combination of X-type twist drill, 2462 rpm spindle speed and 110 mm/min obtained from PSO that can minimize the delamination in the hole entry and exit simultaneously during the drilling process of CFRP.