In order to improve the inspection success rate and velocity of an in-line Automated Optical Inspection (AOI) system of Printed Circuit Boards (PCBs), the solder joints of a PCB are examined. The solder joint images are acquired by a structure illuminator and a 3-CCD color camera. Based on these images, the area features of the conventional types of PCB solder joints such as good, excessive, poor and pseudo are extracted with respect to the key sub-region in the solder joint. Five kinds of feature matrix models of solder joints are presented. A pattern matching algorithm for inspecting the solder joint is developed by maximizing the similarity of the same type of solder joins. To solve the problem of the threshold determined by experiences, a parameter adaptive learning strategy is presented. Finally, 1 040 chip solder joints on a PCB are inspected. Experimental results show that the success rate is as high as 96.5% and the inspection time is 9 s by using the proposed algorithm.This indicates that the proposed algorithm can achieve both a high success rate and inspection velocity.