Abstract: Based on the principle of optical interference and vision detection technique, a machine vision based glass defect inspection system was developed. The target image was acquired, preprocessed and segmented. Its features were extracted and regional parameters were calculated. The defect information such as tin spots, air bubbles and glass ribs was obtained. The experiment indicates that the algorithm is simple, fast and reliable. The system can detect defects down to 0.3mm with maximum processing cycle less than 400ms.