Quality of data is a central concern in epidemiological studies, particularly when using secondary data. This study aims to carry out on reliability and accuracy of stomach cancer mortality data in the city of Rio de Janeiro. A simple random sample of 97 death certificates was obtained from a total of 645 stomach cancer deaths in 1990, and was used for validation. A questionnaire specifically designed to obtain clinical and laboratory data was completed for each case. Two physicians read this questionnaire and completed the new death certificates, allowing analysis of the instrument's reliability. A total of 86 cases were studied as a consequence of 11 losses (11.3%). Reliability analysis of questionnaires revealed 90.7% agreement according to stomach cancer diagnosis (kappa = 0.73). Accuracy was calculated by positive predictive value: 90.7%. Eight cases were discharged after clinical and laboratory revision. The study concludes that mortality data by stomach cancer in the city of Rio de Janeiro are very reliable, and that their level of accuracy is adequate for use in epidemiological studies.