Age of onset is an important variable in Alzheimer research, yet remarkably little is known about its reliability and validity. The present study evaluated alternative methods for establishing age of onset for 297 individuals diagnosed with dementia through the HARMONY study. We compared two informant-based methods: a single question in a telephone screening interview and an in-depth, semi-structured, in-person interview. We then compared informant reports with medical records. The two informant-based methods yielded very similar results, r(297) = .89 (P < .0001). Informant reports were highly correlated with age of diagnosis in medical records, r(155) = .88 (P < 0.0001) for the single question, r(201) =.89 (P < .0001) for the in-depth interview. Age of diagnosis in the medical records lagged consistently 2.9 and 2.7 years behind age of onset from screening and in-depth interviews, respectively. Greater lag was associated with longer time since dementia onset, whether due to recall bias or improved detection of dementia. Lag was not influenced by dementia type, informant's relationship to proband, or proband's age. These findings suggest that informant reports of age of dementia onset are reliable and reasonably valid, even when only a single question is asked within the context of a larger interview.