This paper addresses with the problems in automated visual inspection of automotive instrument cluster dials for which appearance quality criteria are based on human sensibility. By analyzing the experimental human visual sensibility for defects, two types of defect identification method are proposed for pinhole and for particle and scratch. To execute these identification methods in real-time, a pipelined image processor composed of FPGA boards has been newly developed and introduced into the dial manufacturing line as an automatic inspection system. The result of test operation was satisfactory.