Although a number of studies have examined depression risk factors for elderly persons, little attention has been paid to the prediction of individuals at risk. This study constructed a predictive model for discrimination between individuals at a higher risk of depression and normal subjects in Japanese community-dwelling elderly persons, using linear discriminant analysis. Data were collected from 754 non-institutionalized elderly men and women aged 65 years and older living in the community in Japan, using face-to-face interviews in 2002. Stepwise linear discrimination analysis was used to construct a predictive model to select individuals who have a higher risk of depression. The stepwise discriminant analysis selected the five predictor variables (frequent hearing problems, poor appetite, less financial leeway, low emotional support and less subjective usefulness) and yielded a statistically significant function (λ=0.816; χ2=113.0, df=5, p