This study proposed an automatic optical inspection (AOI) system for detection of thin-film transistor (TFT) array defects. gray level co-occurrence matrix (GLCM) and MATLAB regionprops function were used to calculate 53 TFT array defect features, which were inputted into the neural network to train the defect classifier. The images to be inspected were compared with a standard image first, in order to judge whether the TFT array samples have defects. For defective images of a TFT array, the proposed defect classifier can successfully recognize five kinds of defects in the process. According to the experimental results, the defect recognition rate of proposed system is verified to be 83.3%, which can replace manual inspection and reduce the risks of false inspections due to long duration manual work. Moreover, the proposed AOI system can improve testing efficiency and reduce manufacturing costs.