A dot matrix character segmentation method based on machine vision was proposed to address the low recognition rate of dot matrix characters caused by complex backgrounds, local leakage, or blurred characters. First, the rectangular area where the dot matrix characters were located was quickly located from the image using Blob analysis and rotated to the horizontal direction. Then, the rectangular area was segmented into two rows using horizontal grey scale projection, and the characters in each row were segmented using vertical grey scale projection combined with character width features. Finally, the segmented characters were fed into the MLP classifier for recognition. The experimental results show that the segmentation accuracy of the method is 98.7%, which is significantly better than existing segmentation methods, and also has strong recognition generality; therefore, the method has good application value in the recognition of dot matrix characters.