The work aims to improve the glass-bottle defect detection accuracy and ensure production line packaging efficiency. A bottle defect detection method was designed based on machine vision. The overall framework of detection system was briefly introduced. The methods of image segmentation based on maximum entropy, bottle positioning and image feature extraction were respectively discussed. Image features mainly included the perimeter, circularity and relative distance of circle's center. The accurate bottle defect recognition was realized with BP neural network and the bottle damage degree was converted into a specific value. Finally, the experimental verification was carried out. The success rate of the proposed detection method for the damaged bottle was 99%. It had higher detection accuracy for different damage types. The glass-bottle defect detection method based on machine vision can meet the requirements of production line for accuracy and real-time performance.