The work aims to design a method of bottle mouth positioning and defect detection based on machine vision, in order to realize the on-line defect detection of glass bottle. The system structure, including camera, control system and elimination mechanism, was introduced. The image processing algorithms, such as image prepossessing, image seg-mentation, bottle mouth positioning and defect detection, were discussed in detail. Glass bottle image noise reduction was completed by median filtering. Image segmentation was realized by iteration threshold. The bottle mouth was positioned based on the pixel coordinate average method. The radial integral projection and double circumferential scanning were used to detect the bottle mouth defects, such as bottle mouth fracture and damage. Finally, an experimental study was carried out. The experimental results showed that the performance indexes of the proposed algorithm were above 98%, and the accurate, fast and untouched detection was achieved. The system is easy to operate and can meet the requirements of practical use.