Quality control is one of the important issues in the flexible printed circuit board (FPC) making industry. This paper propose a unsupervised defect detection approach based on wavelet packet fame for solving the problem of automatic inspection of Flexible Printed Circuit boards (FPC) gold surfaces. Detection of defects within the inspected texture of FPC gold surfaces is performed first by decomposing the gray level images into subbands using wavelet packet frame, then by extracting corresponding features from some important subbands selected according to their subwindow average energy, finally by reducing feature dimensions with principal component analysis (PCA) and thresholding the result feature images. A number of images of FPC gold surface with a variety of defects have been tested and the experimental results which yielded visually good segmentation confirm the validity of the proposed algorithm.