In the manufacturing process of machine products, it is important to automate the manual visual inspection in detecting microscopic surface defects in order to improve the efficiency and eliminate human errors as well. The basic hardware system consists of a high-resolution camera equipped with a telecentric lens, and a device to change light directions using 60 LED light bulbs. We introduce an accurate automatic system to detect such surface defects based on the novel hardware system and the iterative photometric stereo techniques, which iteratively improve the quality of the estimation of the surface shape. Complex examples are provided to demonstrate the effectiveness of the proposed system.