When the surface defect of a printed-circuit-board (PCB) is detected by a traditional method, it usually has lower detection accuracy for that the traditional method of standard board is built by properties of PCB image itself. According to the accurate requirements of PCB detection, this paper proposes a new detection system. In this system, the layers of PCB photoelectric image were separated by using Gerber file that have been analyzed, then the analyzed Gerber file was corrected by morphology. And a accurate standard board was built. Furthermore, the frequency information in the color image was extracted based on the principal component analysis, and the detection threshold was set according to different information on different layers. By which, the local detection of PCB was completed and detection accuracy was improved. As compared with the traditional method by color stratification, this method improves detection accuracy and detection efficiency of the system. It achieves the detecting probability of 95.1% for micro defects and the detecting time of 1.09 s for a 25 cm×22 cm PCB. The detection efficiency of automated optical inspection system can satisfy the requirement of the PCB for defect inspection in online and real time.