A method, due to S. Winograd, for multiplying matrices that requires fewer multiplications than a straightforward calculation as a result of ‘pre-processing’ the two matrices concerned. This involves storing vectors that are used several times in the calculation.