Studies on risk factors for drug non-compliance have not taken into account the possibility of correlated outcomes. We therefore conducted a study into risk factors for non-compliance by using analysis techniques that adjust for these correlations (longitudinal data analysis). Data were obtained from interviews and pharmacy records in a cross-sectional survey in Amsterdam. The subjects were 157 elderly people aged 70 years or older. Of these subjects, 37 were residents of a home for the elderly, 40 were community-dwelling elderly who needed to be visited regularly by a district nurse, and 80 were community-dwelling elderly who did not need to be visited by a district nurse. Most drugs (78%) were used according to the directions; the remainder (22%) were not used as intended. Odds ratios (95% confidence intervals) for non-compliance for moderate and poor/wrong knowledge of the purpose of a drug as compared with good/correct knowledge were 2.8 (1.2-6.7) and 4.2 (1.5-12), respectively. Drug regimens of two times daily and more than two times daily were associated with odds ratios for non-compliance of 4.5 (1.6-12) and 4.2 (1.7-11), respectively, compared to a regimen of once daily. Compliance increased if a drug was prescribed by a specialist instead of a general practitioner odds ratio 0.1 (0.04-0.4)]. There was no significant relation between compliance and the number of drugs prescribed to a patient, sex, age, living situation, patient group, or perceived effect. This study, which was based on longitudinal data analysis, demonstrates that in elderly people non-compliance with drug therapy is related to the knowledge of purpose of a drug, the complexity of a drug regimen, and the type of prescriber. The positive association between compliance and the number of drugs prescribed found in former studies was not confirmed.