Multiple synthetic discriminant function (SDF) filters has been verified to be a better strategy for distortion invariant optical pattern recognition, especially for the in-plane rotation invariant ones. However, how to select the training sets is very crucial for the filter’s validity. In this paper, referring to the Rayleigh criteria, the distortion tolerance of a single matched spatial filter (MSF) has been tested by using a digital-optical hybrid 4f correlation system. Experimental results are ±5° and ±8% for the rotation and scaling distortion tolerance, respectively. According to the results, an equal correlation peak phase-only SDF algorithm has been adopted to synthesize two SDF filters, in which the training set images have been correspondingly selected with 5° and 10° intervals. The SDF filter constructed with 10° interval training sets has demonstrated as good performance as that one with 5° interval, but with the advantage of reduction by half of the number of multiple SDF filters for 360° full-rotation, which indicated that the double of distortion tolerance selected as the interval of training images is more suitable for real-time correlator recognitions.