Although the predictability of untoward drug effects in humans has improved in recent years, certain new drugs, with new pharmacological mechanisms, still pose a considerable challenge. This holds particularly true for biotherapeutics and their drug-related immune reactions, idiosyncratic drug hepatotoxicity and systemic toxicity. The selection of the 'right' animal models remains crucial; the species selected must be relevant (to humans) and sensitive with regard to three basic variables: pharmacodynamics, pharmacokinetics (including metabolism) and the mechanisms underlying the toxicity in the target human diseases. Furthermore, normal healthy animals might be a poor model in certain cases because the underlying disease in patients can be an important determinant of susceptibility to adverse effects. Therefore, we suggest that, where appropriate, new animal models of human disease (s) are introduced into drug safety assessment.