The construct validity of the 15-item Geriatric Depression Scale (sfGDS) has been assessed in selected populations. The aim of this study was to assess the appropriateness of applying the sfGDS to unselected older inpatients. The main component analysis of sfGDS was performed in 2032 medical inpatients (mean age = 76.3 +/- 8.4). sfGDS did not qualify as a unidimensional test. Three factors explained 47.7% of variance and explored the following dimensions: positive attitude toward life, distressing thoughts/negative judgment about the own condition, and inactivity/reduced self-esteem. The internal homogeneity was poor (Cronbach's alpha = .46). A higher fraction of variance was explained in patients independent in all or dependent in > or = 1 activity of daily living (ADL). In older medical inpatients, sfGDS is not a single construct, which prevents the univocal interpretation of the final score. The higher fraction of explained variance in patients with comparable ADL performance probably reflects the dependency of affective from physical status.