This study examined the temporal stability of the measurement structure of the Center for Epidemiologic Studies Depression Scale (CES-D; Radloff, 1977) in 813 individuals with rheumatoid arthritis. Participants completed the CES-D (Radloff, 1977) on three occasions 1 year apart. Structural equation models and polyserial correlations were used to address methodological limitations of previous studies. Four competing measurement structures were tested with one factor, three factors, four factors, and a single second-order factor underlying the four-factor model. The four-factor and the second-order-factor models provided the best fit at Time 1. When cross-validated at Times 2 and 3, the four-factor and the second-order-factor models remained invariant. Researchers can now more confidently use the CES-D to examine how distress changes in chronic physical disorders.