Optimal cutoff points based on response to the General Health Questionnaire (GHQ) used to identify cases of psychological disorder have serious limitations. We used a latent class model (LCM) to improve case identification, on a national survey data of 999 individuals aged 65 + years, living in Britain. The method suggested three clusters of homogeneous response to GHQ-12, comprising 70% noncases, 24% possible, and 6%, probable cases. The three groups differ in characteristics including objective ones, such as health and subjective such as optimism. The solution seems sensible in recognizing a range of case severity, which will help in applying suitable interventions on mild and moderate cases that are common among older people and are likely to become more serious.