Thus far we have been concerned with the design and analysis of experiments from the point of view of reducing the random errors by arranging the units in groups. We have considered the treatments only as a heterogeneous group within which we can compare any pair of treatment means using a t-test based on the error MS from the analysis. In many experiments the treatments have some structure, so that we may wish to make particular comparisons between and within groups of treatments instead of the simple comparison of all possible pairs of treatments.