Background Depression in older adults is often under recognised despite it being the most common mental health illness in this age group. An increasing older adult population highlights the need for improved diagnostic rates. Brief versions (15 items or less) of the Geriatric Depression Scale (GDS), which are suitable for busy clinical practice, could improve detection rates. Objective Our aim is to establish the diagnostic accuracy of brief versions of the GDS. Methods Twelve electronic databases of published and unpublished literature were searched. Study selection was in accordance with predefined inclusion and exclusion criteria. A recognised gold‐standard diagnostic instrument was used as a comparator against data pertaining to the use of a brief version of the GDS in an older adult population. The QUADAS‐II was utilised for quality assessment. Narrative analysis and, where possible, meta‐analysis were performed. Results Thirty‐two studies were identified that provided diagnostic data regarding seven brief versions of the GDS (1, 4, 5, 7 8, 10 and 15‐item versions). Pooled sensitivity was 0.89 (95% confidence interval (CI) 0.80–0.94), and specificity was 0.77 (95% CI 0.65–0.86) for the GDS‐15 at the recommended cut‐off score of 5. Meta‐analysis of other brief versions was not possible because of an insufficient number of studies with standardised items. Conclusions Results suggest the possibility of selective reporting of cut‐off scores, and therefore, findings should be approached cautiously. Studies should report all cut‐off scores, and all brief GDS versions should be compiled of standardised items. Copyright © 2016 John Wiley & Sons, Ltd.