This paper deals with three problems related to comparing independent treatment groups to a control. The first goal is to extend Dunnett's solution to two‐way designs. Let μjk(j= 1, …,J;k= 1, …,K) be the means ofJKindependent normal distributions. Percentage points of the multivariatetdistribution are supplied that provide two‐sided 1–α simultaneous confidence intervals for μjk– μjk(j= 1, …,J– 1;k= 1, …,K) when the variances and sample sizes are equal. Next, approximate methods for handling both unequal variances and unequal sample sizes are suggested and studied via simulations. A new two‐stage procedure is also described where the probability coverage is at least 1–α and the length of the confidence intervals are fixed in advance by the experimenter, even when the variances are unequal. Using real data, implications of the new results are illustrated for situations where a researcher is interested in performing multiple comparisons with the ‘best’ treatment group, where ‘best’ is defined to be the treatment group with the largest mean. In particular, the results in this paper can be used to determine whether the most effective of several treatments was correct