Autopsy can represent a key instrument in auditing clinical diagnosis performance, on which in turn the accuracy of diagnoses, as used in a variety of epidemiological investigations, depend. This audit can be implemented through clinico-pathological surveys which require both accurate pathological examinations and validity of study design and analysis. To this end, estimates of sensitivity and specificity of the clinical diagnosis, using the autopsy diagnosis as a yardstick, should be obtained, which can however be severely distorted by factors such as nonrandom selection of cases for autopsy or by unrecognized errors in post-mortem diagnosis. Such distortion may be minimized by (a) estimating the likely magnitude of errors in postmortem diagnosis, (b) specifying standard conditions for performing autopsies and (c) ensuring an unbiased sample of moderate size rather than a large biased sample. Considerable improvement in the validity of clinico-pathological surveys as carried out up until now is possible, and there is room for research to provide the necessary information (e.g. on necropsy diagnosis variability and feasible sampling schemes).