Objective: This study aims to investigate if a brief version (four items) of the Geriatric Depression Scale works equally well with mildly demented as with nondemented patients in young-old and old-old persons, and to compare its diagnostic performances with those of the 15- and the 30-item version of the scale. Method: Four hundred and forty-two older persons were given a GDS interview and received an independent psychiatric evaluation. Receiver operating characteristic curves were separately plotted for the young-old demented, old-old demented, young-old nondemented, and old-old nondemented. Results: The four-item version yielded comparable performance to the 15- and the 30-item version of the GDS, regardless of age and dementia status. It is reasonably robust to the effects of age and mild dementia, whether alone or in combination. Nonetheless, among the old-old demented, only a third of those tested positive were actually positive, but this problem was not specific to the four-item version. Conclusions: The four-item version can be used in lieu of the longer versions to conserve clinical and research resources where appropriate, and a cutoff of 1/2 can be applied across the board. Little, if any, information is lost when this brief version is used instead of the longer ones.