A statistical design for experiments that removes from experimental error the variation from two sources that may be identified with the rows and columns of a square. The allocation of experimental treatments is such that each treatment occurs exactly once in each row and column.