Objective Self‐neglect is an imprecisely defined entity with multiple clinical expressions and adverse health consequences, especially in the elderly. However, research has been limited by the absence of a measurement instrument that is both inclusive and specific. Our goal was to establish the psychometric properties of a quantitative instrument, the Abrams Geriatric Self‐Neglect Scale (AGSS). Methods We analyzed data from a 2007 case–control study of 71 cognitively intact community‐dwelling older self‐neglectors that had used the AGSS. The AGSS was validated against two “gold standards”: a categorical definition of self‐neglect developed by expert consensus; and the clinical judgment of a geriatric psychiatrist using chart review. Frequencies were examined for the six scale domains by source (Subject, Observer, and Overall Impression). Internal consistency was estimated for each source, and associations among the sources were evaluated. Results Internal consistency estimates for the AGSS were rated as “good,” with the Subject responses having the lowest alpha and omega (0.681 and 0.692) and the Observer responses the highest (0.758 and 0.765). Subject and Observer scores had the lowest association (0.578, p < 0.001). Using expert consensus criteria as the primary “gold standard,” the Observer and Overall Impression subscales were “good” at classifying self‐neglect, while the Subject subscale was “fair.” Conclusions The AGSS correctly classified and quantified self‐neglect against two “gold standards.” Sufficient correlations among multiple sources of information allow investigators and clinicians to choose flexibly from Subject, Observer, or Overall Impression. The lower internal consistency estimates for Subject responses are consistent with self‐neglectors' propensity to disavow symptoms. Copyright © 2017 John Wiley & Sons, Ltd.