Objective Self‐report measures of depression are highly important tools used in research and in various healthcare settings for the diagnosis of different levels of depression. The Geriatric Depression Scale (GDS) is the first and the most popular scale used to screen for late‐life depression. It is endorsed by the Royal College of Physicians and the British Geriatric Society (1992). The purpose of the present research was to investigate whether scores on the GDS15 capture depressive mood (i.e. trait depression), depressive affect (i.e. short‐term depressive state), or both. Methods For this purpose, a trait–state model (stable trait, autoregressive trait, and state model) was applied to GDS15 scores obtained at four time points over a 6‐year period among a sample of community‐dwelling older persons (N = 753). This model allows decomposing the GDS15 scores into three different variance components: stable trait variance, autoregressive trait variance, and state variance. Results Our findings revealed a general pattern of a major proportion of stable trait (69%) and autoregressive trait (22%) variance and a very smaller amount of state variance (9%) in the GDS scores across 6 years. Age and gender (i.e. being female) were shown to be positively linked to more stable trait variance. Conclusions Depression, as assessed with the GDS15, should be regarded as a relatively stable and enduring trait construct, reflecting a stable core of a person's depressivity. The negligible amount of state elements in the variation of the GDS15 scores provides evidence that changing the context will not be enough to cause significant changes in depressive symptoms. Copyright © 2016 John Wiley & Sons, Ltd.