The invention relates to an online visual detection method for a motor control panel. The method includes S1) a step of identifying a defect part, and S2) a step of marking the defect part. Multi-angle images of the motor control panel are acquired through cooperation of first and second image collecting devices, thus avoiding obstruction of a large-power device and making detection more accurate. After the defect is identified, the defect part can be marked, thus facilitating further artificial identification.