In this paper, an inner-product array processor for the associative retrieval problem is presented. First, the algorithm and architecture of the array processor design are discussed. Then an optical implementation scheme is proposed. The matrix model of the associative memory is adopted. In this model, if one of the M vectors is to be reliably recalled, the dimension of the vectors, N, must be much larger than M. By taking advantage of this fact, our result offers a factor of 1/1â- 17â€"M- saving on the matrix elements. More significantly, real-time inputting and updating of the matrix elements can be potentially implemented with existing space-variant holographic elements and recently discovered liquid crystal television spatial light modulators.