OBJECTIVE: Diagnostic classification and reliability are critical in genetic linkage studies of schizophrenia and bipolar disorder. To establish an optimal diagnostic procedure, the authors drew 13 methodological elements from 38 major linkage studies and workshop reports. They determined reliability for a consensus best-estimate diagnostic method based on these 13 features. METHOD: Each of 59 subjects from several large multiplex pedigrees, densely affected by either schizophrenia or bipolar disorder, received a best-estimate diagnosis from unblind diagnosticians in the field and also from a panel of four research psychiatrists who were blind to the proband's and relatives' clinical status. The best estimate was based on personal diagnostic interviews, all available medical records, and family history data. RESULTS: The diagnostic concordance between the field team and the blind psychiatric board yielded 78% to 90% agreement for the whole sample (kappa = 0.83-0.88) and 71% to 87% agreement for the subjects given field diagnoses (kappa = 0.76-0.83). The diagnoses made by the unblind field diagnosticians were biased toward a greater severity (or certainty) level in the diagnostic hierarchy (schizophrenic or bipolar) and more consistency with the most prevalent diagnosis affecting the pedigree. CONCLUSION: Since several previous linkage studies used diagnoses made by diagnosticians who were not blind to the status of the probands and the relatives or did not use a consensus best-estimate diagnosis, further reliability studies of different aspects of the best-estimate method and of its effect on linkage studies are needed. Such research is imperative given the serious impact of diagnostic misclassifications on genetic linkage results.