The iterative proportional fitting (IPF) algorithm is simple to use for fitting hierarchical log-linear models of any size whether their maximum likelihood estimates are given in a closed form or not. This paper shows that the algorithm can be extended toward a class of nonhierarchical log-linear models.