Diagnosis of hereditary non-polyposis colorectal cancer (HNPCC) is currently based on phenotypical analysis of an expanded pedigree. Diagnostic guidelines ('Amsterdam criteria') proposed by the International Collaborative Group on HNPCC are often too stringent for use with small families. There is also the possibility of false-positive diagnosis in large pedigrees that may contain chance clusters of tumours. This study was conducted to determine the effect of family size on the probability of diagnosing HNPCC according to the Amsterdam criteria. A total of 1052 patients with colorectal cancer were classified as HNPCC or non-HNPCC according to the Amsterdam criteria. Associations between this diagnosis and the size of the first-degree pedigree were evaluated in logistic regression and linear discriminant analyses. Logistic regression showed a significant association for family size with the Amsterdam-criteria-based HNPCC diagnosis. Linear discriminant analysis showed that HNPCC diagnosis was most likely to occur when first-degree pedigrees contained more than seven relatives. Failure to consider family size in phenotypic diagnosis of HNPCC can lead to both under- and overestimation of the frequency of this disease. Small pedigrees must be expanded to reliably exclude HNPCC. Positive diagnoses based on assessment of eight or more first-degree relatives should be supported by other clinical features.