An algorithm is presented for filling in least squares estimates of m missing values in any analysis of variance design. The method is non-iterative and requires only those subroutines already in use ...