The two-sided bivariatet-test for simultaneous comparison of two treatments with a control is studied. The problem of allocating a given number of observations to the treatments and the control is discussed, and it is concluded that, in the absence of preferences, a general purpose rule would be to choose the sample sizes equal. A table is presented which gives the sample size required to ensure that the null hypothesis will be rejected with at least a prescribed probability, given the significance level and the standardized differences between the treatments and the control. Similar tables are provided for use when the object is to ensure that both treatments means will be found significantly different from the control mean with at least a given probability, or to ensure that a specified treatment will be found significant with at least a given probability.