Trace is the important composition structure of printed circuit board, which connects the devices, it is also the module that taking up the highest proportion, thus, it is the major testing target of quality assurance. The sunken defects proposed in this paper is a width abnormity defect on the FPC trace, which would cause the latent open circuit defect and affect the electrical function of circuit. The problem of flexible deformation and width difference make the FPC trace detection harder. Therefore, this paper proposed a detection scheme combined with linear masks and circle distribution characteristic. Firstly, this scheme preprocessed the acquired FPC image, divided the trace into several sub-regions and obtained the line width values of each trace transverse section. Then the line width sequences were searched with the linear difference template of gray scale. Thus, the sunken defects alternative positions were located. Lastly, the circle distribution characteristic is defined to identify the real defect areas from the alternative regions acquired from the previous step. Thus, the detection of sunken defects on the FPC trace was accomplished. The algorithm was tested in the self-built image database, which shows the better detection performance than the other typical algorithms.