The purpose of this study was to examine the optimal scoring scheme (category use), unidimensionality, item fit, and redundancy of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) in a cohort of Hong Kong Chinese stroke survivors. At 3 months after the index stroke, a research assistant administered the IQCODE to relatives of 284 Chinese patients with acute stroke who were consecutively admitted to a general hospital. A psychiatrist, who was blinded to the IQCODE scores, interviewed all 284 patients and made DSM-IV diagnosis of dementia, which served as the benchmark for judging the performance of IQCODE in screening dementia. The results suggest that the optimal IQCODE scoring scheme has 2 rather than the original 5 categories. Although the IQCODE was unidimensional overall, there was evidence of item redundancy, thus indicating that a shortened version is desirable.