In large studies and under field conditions common to epidemiological research, factors outside of and inside the laboratory can introduce misclassification of genetic susceptibility markers. Few reports have been made on the accuracy of genotyping individuals using DNA extracted from frozen urine that was stored for approximately 20 years. This study was performed to determine the reproducibility and accuracy of N-acetyltransferase 2 (NAT2) genotyping by RFLP analysis using DNA from stored urine. To obtain long-term frozen urine and blood samples from the same person, the databases of two large prospective studies were linked by name and date of birth. Six polymorphisms within the coding region of NAT2 were determined in 65 urine and blood samples after which, genotypes and imputed phenotypes (rapid, slow) were derived. To test reproducibility, all of the six polymorphisms were determined twice in 47 urine-blood pairs. Reproducibility of imputed phenotypes was 91.5% in urine samples and 97.9% in blood samples. To test accuracy, results for all six polymorphisms were compared between urine and blood DNA. All of the kappa's were at least 0.85 except one. Identical results for all six polymorphisms were seen in 78.5% of urine-blood pairs. Taking blood samples as a reference standard, rapid acetylators were classified as rapid in 97% of subjects (95% confidence interval, 90-100%), and slow acetylators were classified as slow also in 97% of subjects (95% confidence interval, 91-100%), when using urine. This study shows that stored urine samples can be used for DNA genotyping in large cohort studies, when blood samples are not available.