Hereditary nonpolyposis colorectal cancer (HNPCC) is an autosomal dominant syndrome caused by germline mutations of the mismatch repair (MMR) genes. Only a few studies have taken into account the selection of families tested for these mutations in estimating colorectal cancer (CRC) risk in carriers. They found much lower estimates of CRC risks than previous ones, but these estimates lacked precision despite the large number of families. The aim of this study was to evaluate the efficiency of the 'genotype restricted likelihood' (GRL) method that provides unbiased estimates of risks whatever the ascertainment process of families, and to estimate CRC and endometrial cancer risk for carriers of the MMR genes. Efficiency of the GRL method was evaluated using simulations. Risks were estimated from a sample of 36 families diagnosed with HNPCC and carrying a mutation of MSH2 or MLH1, ascertained through a cancer family clinic in Lyon (France). The efficiency of the GRL method was found to be strongly dependent on the proportion of family members tested. By age 70 years, CRC risk was estimated at 47% (95% confidence interval: 12-98%) for men and 33% (95% confidence interval: 24-54%) for women. The endometrial cancer risk was only 14% (confidence interval: 6-20%). As methods allowing for the selection of families lack efficiency, large-scale family studies should be undertaken and data should be pooled to provide reliable and precise estimates of risks for an optimal familial management.