We examined agreement between data abstracted from medical records and interview data for patients with dyspepsia admitted to hospital for endoscopy, to determine the extent to which health records could be used to validate self-reports of dyspepsia and the management of this condition. Results from the sample of 220 patients showed that there was poor agreement between data sources for information about duration of dyspepsia (k=0.34) and previous barium meal examination (k=0.34). Patients reported significantly longer dyspepsia histories (Wilcoxon sign test Z=4.13, p<0.0001) and significantly more barium meals (sign test Z=8.43, p<0.0001) than were documented in their records. There was also disagreement between data sources regarding the number of drugs taken before and after endoscopy (k=0.28 and k=0.31, respectively). Where there was disagreement for number of drugs there was no significant difference in the direction of the disagreement. There was moderate agreement regarding the name of pre-endoscopy medication (k=0.55) and substantial agreement for the name of medication used post-endoscopy (k=0.62). There was very poor agreement regarding diagnosis. The medical record was the gold standard for this information. Choice of data source, medical records or self-reports, will in many instances provide significantly different results and it is likely that this may also be true for other variables of interest to researchers. Thus in the case where no gold standards are available researchers need to consider carefully the implication of choice of data source on their results.