RATIONALE, AIMS AND OBJECTIVES: We investigated the validity of self-reported admission data compared to administrative records in a clinical trial. METHOD: In the randomized KORINNA study (ISRCTN02893746), hospital admission data were collected in telephone interviews with 273 elderly patients quarterly over a 1-year period and thereafter annually over a 2-year period. Data were compared with administrative records and discharge letters. Mixed models were used to investigate if recall period and individual characteristics influence validity. RESULTS: Specificity (>99%) and sensitivity (94%) of self-reported data did not differ for different recall periods (3 months vs. 12 months). The differences between self-reported and registered inpatient days were not statistically significant. Having regard to all the admissions within the time period of last interview and dropping out, the bias was up to 40% underestimation. The chance of disagreement was significantly smaller [odds ratio (OR) of misremember an admission = 0.596, P = 0.049, confidence interval (CI) = 0.355 to 1.00; OR of misremember length of stay = 0.521, P = 0.002, CI = 0.344 to 0.789] for 3-month periods, but this was primarily driven by number of admissions within the recall period. Individuals with better health and longer stays had a significantly smaller chance of disagreement. CONCLUSIONS: The bias within one year was not influenced by applying various recall periods, although the probability of correctly self-reported single hospital admission was higher using a recall period of three months. It can be recommended that lengthened recall periods of 12 months are appropriate for gathering self-reported hospital admission data in elderly people with myocardial infarction.