In recent years wood strength researchers have begun to replace experimental unit allocation via random sampling with allocation via sorts based on nondestructive measurements of strength predictors such as modulus of elasticity and specific gravity. Although this procedure has the potential of greatly increasing experimental sensitivity, as currently implemented it can easily reduce sensitivity. In this paper we discuss the problem and we present solutions. Given the existence of nondestructive measurements of strength predictors, our methods can be used to reduce sample sizes. We have written a public domain computer program that implements the methods.