An optical associative processor that employs variable nonlinearity in the filter plane of the conventional-filter-based optical associative memory is described. It is shown that for severe compression types of nonlinearity, the performance of the associative processor in the areas of retrieval quality, convergence rate, and light efficiency is improved significantly. The nonlinearity in the filter plane is implemented using a nonlinearly transformed filter to store the associative images such that nonlinear correlations between the input image and the stored images are obtained. Two methods are described to construct the nonlinearly transformed filter. One method applies uniform nonlinear transformations to the filter function containing all associative signals. The second method applies controlled nonlinear operations on the filter function. Analysis and computer simulations are presented to illustrate the importance of applying nonlinear transformations in the filter plane of the optical associative processor.