Over the last two decades animal models, mediators and drugs have produced a mass of experimental data difficult to relate to the epidemiological and clinical side of psychiatric disorders. Antidepressant drugs are chosen as a model case to describe a more general scene: far from being a specific tool--a guide--in a complex situation, these drugs are acquiring a role which must be interpreted with caution (a caveat), in view of ample clinical evidence of responders to placebo, responders to any drugs, responders to many drugs, etc. Studies of biochemical descriptors of drug action as a marker of the disease and its outcome have drawn puzzling pictures, often contradictory and unstable in terms of the populations to whom they can be applied. Controlled clinical trials with antidepressant drugs over the last ten years have persisted in looking for short-term pharmacological effects rather than the medium- or long-term impact of medication in large populations. To establish a positive role for antidepressant drugs, they must be studied in a "natural" context where depressed patients are treated with all necessary follow-up.