Failure to obtain convincing results in psychiatric genetics can partly be attributed to the fact that progress in molecular biology and genetic epidemiology has not been followed by an equivalent development in phenotypic description. Instead of relying entirely on classical nosological approaches, we argue that identifying more homogeneous forms of diseases through a'candidate symptom approach' among affected subjects and an endophenotype approach that identifies sub-clinical traits among non-affected relatives might yield better results. Examples where these strategies have already been fruitful when applied to complex diseases are presented in this review. Focusing on vulnerability traits might stimulate the redefinition of traditional psychiatric syndromes and help to bridge the gap between clinical and experimental approaches.