In the search for novel treatments for psychiatric disorders, many compounds that have shown promising pharmacological properties in disease models have failed to induce benefit in patients. There is good reason to believe that the preclinical approaches routinely used in drug discovery often provide an overly optimistic picture of clinical potential. Here we discuss some of the factors that we believe lead to erroneous decision-making, including: false interpretations of the behavioural significance of drug effects in the model species; fundamental flaws in aspects of experimental design and analysis; and misconceptions about the criteria that need to be applied before a model can be said to be validated. Only by focusing on well-constructed biological hypotheses of drug action in conjunction with reliable neurochemical, electrophysiological and behavioural assays that can be demonstrated to engage clinically relevant brain circuits will the chances of clinical success be improved. As psychiatric disorders come to be viewed less descriptively and more mechanistically as developmental disorders in brain circuits, incorporating biomarkers – measured biological variables that can indicate a normal or abnormal biological etiological process – will become the essential key to improving model development and validation, and target assessment and refinement.