Late-life depression is a major public health concern, associated with poor health outcomes, including doubling of dementia risk. Psychiatric evaluation is impractical in large epidemiological studies, which instead typically rely on self/informant reports, which are subject to various biases (stigma, recall). Few studies have addressed level of agreement between sources. This study examined associations between these sources and assessed whether subject and informant variables moderated these associations. In a population-based study of dementia in Cache County, Utah (2002-5), 1,480 subjects completed an in-depth clinical assessment (CA). Major depression was assessed via the self-report Patient Health Questionnaire-9 (PHQ-9) and informant-rated Neuropsychiatric Inventory (NPI-CA). One hundred forty-eight subjects with cognitive impairment also completed a psychiatrist’s examination, including the self-report Geriatric Depression Scale (GDS), the informant-rated NPI (NPI-MD), and the physician’s clinical rating (PCR). Bivariate correlations were modest: NPI-CA versus PHQ-9 (r = .26), NPI-MD versus GDS (r = .20), GDS versus PCR (r = .22), NPI-MD versus PCR (r = .45). Kappa statistics and logistic regression models indicated that the NPI-CA predicted the PHQ-9 moderately (ϰ = .08, p <.001; OR = 3.1, 95% CI: 1.5 to 6.1). Results also indicated that the GDS did not significantly predict the PCR (ϰ = .10, p > .05; 95% CI: 0.7 to 11.2) nor the NPI-MD (ϰ = .01, p > .05; 95% CI: 0.6 to 6.3), and that the NPI-MD predicted the PCR moderately well (ϰ = .35, p < .001; OR= 11.1, 95% CI: 2.6 to 48.3). CA-NPI predicted the PHQ-9 for cognitively normal subjects (ϰ = .13, p < .001; OR = 10.1, 95% CI: 1.9 to 52.6) but not for subjects with mild impairment (ϰ = .01, p > .05; 95% CI: 0.4 to 4.3) nor dementia (ϰ = .14, p > .05; 95% CI: 0.9 to 7.8). No other variables moderated these associations. Results suggest the importance of cognitive assessment when measuring late-life depression via self-report.