A recognition method of pattern feature is provided, where a recognition result thereof is applied to optical proximity correction, the method includes: providing a plurality of reference images with a reference pattern feature; recognizing and classifying the reference images by an image recognition device, and storing the recognition result; comparing the image with the actual pattern feature with the stored recognition result by the image recognition device to recognize and classify the image with the actual pattern feature; and calculating an angle feature value and/or a distance feature value of the actual pattern feature by the image recognition device according to a classification result to obtain the recognition result of the pattern feature.