The invention discloses an automatic fluorescent magnetic powder defect recognition method based on Laws texture filtering. The method can be used for crack defect detection on the surface and near the surface of a ferromagnetic device and solves the problem of reflection of a metal surface during an actual image acquisition process. Particularly, local texture energy information is calculated through a template, gray variation information can be acquired, and feature images are obtained; the obtained seven-channel feature images are filtered and put to a Gaussian mixture model for classification and recognition; normal areas are obtained after classification; area subtraction is used for a to-be-detected magnetic mark image to obtain a defect area, and pixel statistics is then carried out on the defects; and if the pixel number in the defect area is higher than a certain threshold, the workpiece is regarded as a reject. Compared with the currently-used manual magnetic mark defect recognition, the method of the invention improves the detection efficiency, damages to a human body, particularly, human eyes by fluorescent magnetic powder detection can be reduced, and the automatic degree of the detection device is greatly improved.