In this paper, an optical pattern recognition system based on Vander Lugt correlator architecture is proposed. The proposed system uses the fractional Fourier transform instead of the Fourier transform in Vander Lugt correlator system. The filter function is designed with error back-propagation algorithm, which is popular in artificial neural networks. I develop learning rules for the presented system in both amplitude and phase modulation filters. The system is demonstrated in computer simulations and optical experimentations with some pattern recognition problem involving binary characters.