OBJECTIVES: This study was conducted in order to evaluate reliability and diagnostic validity of the CIDI. METHODS: 64 patients with DSM-IV dementia and 90 elderly patients with other mental disorders entered this study. Inter-subtest Pearson's correlations, corrected subtest-total correlations and Cronbach's alpha were computed for reliability while correlations of the total scores on CIDI with the performances on other scales relating to dementia, sensitivity and specificity, area under curve using the receiver operating characteristic curve, and group comparison were used for the assessment of validity. RESULTS: The CIDI's inter-subtest correlations ranged from 0.596 to 0.813, with a mean of 0.707. The corrected subtest-total correlations were between 0.759 and 0.890. Cronbach's alpha of the subtests was 0.941 and the Cronbach's alpha were between 0.928 and 0.948 when any subtest was deleted. In the demented subjects, the correlation coefficients of the total CIDI score with the performances on S-SDQ, IADL, BADL, BDRS, and MMSEK were around 0.5. Area under the receiver operating curve of CIDI was 0.933 with the standard error of 0.021(95% CI, 0.891-0.975). At the CIDI cutoff point of 65/66, sensitivity and specificity were 95.3% and 85.6%, respectively. The corresponding values of MMSEK were 0.930, 0.020 (95% CI, 0.891-0.969), 22/23, and 89.1% and 85.6%. The demented scored much poorer in all CIDI subtests than the nondemented mental patients, and the differences were highly significant. CONCLUSIONS: The CIDI was satisfactory regarding to reliabilities and diagnostic validities. The CIDI and MMSEK are thought to be reciprocal in assessment of cognitive functions: CIDI is preferable for clinical assessment of cognitive dysfunction and MMSEK for as a screening test.