Early dementia can be difficult to diagnose in older persons from culturally and linguistically diverse (CALD) backgrounds. The Folstein Mini-Mental State Examination (MMSE), the General Practitioner Assessment of Cognition (GPCOG) and the Rowland Universal Dementia Assessment Scale (RUDAS) were compared in 151 older, community-dwelling persons. Receiver operating characteristic (ROC) curve analysis was used to evaluate diagnostic accuracy, while logistic regression was used to evaluate the influence of age, gender, CALD status and years of education. All three instruments were equally accurate in predicting dementia (ROC area under curve 0.92-0.97, p > 0.05 for all comparisons). At the recommended cut-offs, the RUDAS was best for ruling in dementia (positive LR = 8.77), while the GPCOG was best for ruling out dementia (negative LR = 0.03). All three instruments were influenced by concomitant depression. Whereas the MMSE was influenced by CALD status, the RUDAS and GPCOG were not. While the GPCOG combines participant and informant data, the RUDAS is a stand-alone measure specifically designed for, and validated in, multicultural populations. [ABSTRACT FROM AUTHOR]