The prerandomization design due to Zelen (1979, New England Journal of Medicine 300, 1242-1245), for comparing an experimental treatment with a standard, is studied and compared with the traditional randomization scheme. Two models for estimating the treatment effect are considered: one applies when the response variables are continuous and normal, and the other when they are binary. Relative efficiencies are computed and lower bounds, expressed in terms of the probabilities of consent under the two schemes, are given.