In this paper we derive and present critical points for pre-tests in regression using a minimum relative risk criterion. We use the same type risk functions as Sawa and Hiromatsu [8] who, in a recent paper in this journal, derived pre-test critical values using a minimax regret criterion. Since James-Stein type estimators can be shown to dominate any pre-test estimator for the risk functions used here and in [8], no normative claims are made for the critical values we give. However, the use of pre-testing procedures continues in practice and the results given here, contrasted with other results, add to information about the character of costs and returns to such practices. RECENTLY, JUDGE and his associates have derived risk functions and matrices for pre-test estimators in different contexts, and have called attention to the dominance of James-Stein type estimators where continuous weight functions of the tests statistic are used to average the restricted and unrestricted estimators (see [1, 2, and 7]). Sawa and Hiromatsu [8] obtained minimax regret critical values for pre-test estimators where they define regret as the difference between pre-test risk and the minimum of risk for unrestricted and restricted least squares over the range of the noncentrality parameter in the probability density function on the test statistic. They tabulated their minimax regret critical value for the case of a single restriction. Brook [3], with somewhat different scalar risk functions than Sawa and Hiromatsu, derived minimax regret critical values for multiple restrictions, and examined the effect of non-orthogonal X matrices for risk functions that concentrate on the parameter space of the regression coefficients themselves rather than on conditional forecasting of the dependent variable. In contrast to Sawa and Hiromatsu and Brook, the pre-test critical values that we derive are more favorable to unrestricted least squares. For example, using the same risk functions as Sawa and Hiromatsu or Brook, minimum risk of the pre-test estimator leads to opting for ordinary, unrestricted least squares unless the number of restrictions is five or greater, no matter the (finite) number of regressors and observations. That is, our optimal critical value for a pre-test is zero unless the restrictions imply a reduction of five or more in the parameter space. In contrast, even with only one potential restriction, Sawa and Hiromatsu find that the minimax critical value for a pre-test is about 1.8, depending upon degrees of freedom. However, when the number of restrictions is very large, our results are very close to the findings by Brook. That is, critical values are about 2.0 for a large number of potential restrictions for either the minimax regret or the relative minimum risk criteria.