Health technology assessments propose to study the differential impact of health interventions in a complex care system which is characterised by the multitude of individual behaviours and the diverse nature of the institutions involved. Current systems for data collection lend themselves poorly to this rigorous analysis of efficacy of treatments in the actual situations where they are used. Randomised trials endeavour to neutralise any parasitic interference which could compromise testing for a causal relationship between the treatment administered and the result obtained. Their methodology which establishes the term ceteris paribus in the principle of good practice lends itself poorly to an analysis of individual behaviour. Observational studies are start from actual treatment situations to describe them as reliably as possible. By definition, however, these assume that the natural course of events is not deviated by any intervention. The absence of an experimental plan increases the likelihood of bias and makes it more difficult to test for causal relationships. They lend themselves poorly to testing for incremental efficacy. The two instruments to be preferred are decisional analysis and quasi-experimental studies. Decisional analysis help to avoid the problems of external validity associated with randomised clinical trials by associating parameters which are extracted from data obtained from everyday practice. Quasi-experimental studies or pragmatic trials are based on the reality of behaviour of the prescriber and his/her patients; their impact on efficacy, quality of life social costs of the disease and of treatments may be identified under normal conditions of use.