SUMMARY The conventional entry test for a new variable in stepwise regression is not theoretically correct, but is used because the necessary exact distributions are not known. Some distributional results which throw light on the situation are given.