The last 30 years have seen a significant growth in health service evaluation literature in the field of mental health. Much of the best of this research has taken the form of random control trials of an experimental service against "standard' or generic care. Interpretation of these results has ben hampered by incomplete and inconsistent approaches to describing the experimental services, and often inadequate or completely non-existent characterization of the control service. In addition, the health care system in which the treatment programme is embedded is known to have important consequences for outcome. A proposal is advanced for a structured service description as a routine component for health services research reporting. Two worked examples are given, and a number of areas of ambiguity are highlighted.