This paper describes a method of reducing a general matrix to codiagonal form by using a sequence of elementary transformations, and identifies the second stage of the reduction with the Lanczos transformation. Some practical problems of computation are discussed.