This work focused on refining the Cognitive Abilities Screening Instrument (CASI) by selecting a clinically significant subset of tests, and generating simple and useful models for dementia screening in a cross cultural populace. This is a retrospective study of 57 mild-to-moderately demented patients of African-American, Caucasian, Chinese, Hispanic, and Vietnamese origin and an equal number of age matched controls from a cross cultural pool. We used a Knowledge Discovery from Databases (KDD) approach. Decision tree learners (C4.5, CART), rule inducers (C4.5Rules, FOCL) and a reference classifier (Naive Bayes) were the machine learning algorithms used for model building. This study iden- tified a clinically useful subset of CASI, consisting of only twenty Mini Mental State Examination (MMSE) attributes—CASI-MMSE-M, saving test time and cost, while maintaining or improving dementia screening accuracy. Also, the machine learning algorithms (in particular C4.5 and CART) gave stable clinically relevant models for the task of screening with CASI-MMSE-M. …