Abstract: For the inherent nonlinearity of position sensitive detector (PSD),a PSD nonlinear correction method based on backpropagation (BP) optimization algorithm was proposed.Based on the traditional Newton algorithm , the relevant principle of LevenbergMarquardt algorithm was deduced, that is the BP optimization algorithm.The Matlab software was used for programming, the network utilized the structure having 2 hidden layers,and the numbers of neurons for them were 40 and 30,respectively,the max training time was 500,and finally the results were computed by sim function and outputed through simulated network.Through verified by experiment, the network output error is almost within 0.001 mm, of which the maximum error is less than 0.003 mm, achieving the PSD nonlinear correction.