In this paper four principal topics are addressed: (a) the policy and political contexts in which demands arise for cost information; (b) the nature and phasing of those demands; (c) the basic rules of empirical costs research for meeting those demands; and (d) concomitant implications for the design, execution and interpretation of their research. Mental health care policy or practice changes which ignore costs, or which embody cost information without obeying or recognizing the four basic rules, can only be of dubious validity, or can only be used to answer a limited range of questions. But, as the illustrative studies show, it need not be an horrendous, or ideologically compromising or scientifically complex task to add a cost dimension to the evaluation of mental health services. There are enough examples in the literature of bad costs research to demonstrate that it is not as simple as some people think, but there are also enough examples of good research to encourage further attempts.