In this paper, we present a unified approach to evaluating the relative expressive power of process calculi. In particular, we identify a small set of criteria (that have already been somehow presented in the literature) that an encoding should satisfy to be considered a good means for language comparison. We argue that the combination of such criteria is a valid proposal by noting that: (i) the best known encodings appeared in the literature satisfy them; (ii) this notion is not trivial, because there exist encodings that do not satisfy all the criteria we have proposed; (iii) the best known separation results can be formulated in terms of our criteria; and (iv) some widely believed (but never formally proved) separation results can be proved by using the criteria we propose. Moreover, the way in which we prove known separation results is easier and more uniform than the way in which such results were originally proved.