To estimate the agreement between interview-ascertained medication use and pharmacy records among the population aged older than 50 years, and to identify patient-level predictors of discordance. The Irish Longitudinal study on Ageing is representative of community-dwelling adults aged 50 years and older in Ireland. Interview-ascertained medication data from 2,621 participants were linked to pharmacy dispensing records. The kappa statistics measured the agreement between the two sources for 19 therapeutic classes. Logistic regression assessed the effect of patient-level characteristics on survey under- and overreporting of regularly dispensed medications. Agreement was good or very good (κ=0.64-0.86) for 15 medication classes, and moderate or poor for antiinflammatory and antirheumatic products (κ=0.54), analgesics (κ=0.50), psycholeptics (κ=0.59), and ophthalmologicals (κ=0.37). Not reporting an indicated health condition, less frequent dispensing, older age, and more medications regularly dispensed were associated with survey underreporting, but results varied by therapeutic class. Memory and cognition were not associated with discordance. Ascertaining medication use via patient interview seems a valid method for most medication classes and also captures nonprescription and supplement use. However, medications applied topically and as needed may be underreported. The source of medication data should be carefully considered when performing pharmacoepidemiological studies. Copyright © 2013 Elsevier Inc. All rights reserved.