This paper describes an iterative procedure for obtaining maximum likelihood estimates of the parameters of a generalized regression model when direct maximization with respect to all parameters is difficult. A proof of convergence and some interesting applications are provided.