The reliability and validity of the Cognitive Impairment Diagnosing Instrument (CIDI) were studied in 67 nursing home elderly subjects and 251 elderly psychiatric patients. Its possible highest score is 77 and covers 10 subscales: short-term memory, long-term memory, concentration/calculation, abstract thinking, judgement, memory registration, higher cortical functions, orientation in time, orientation in place and object naming. Test-retest correlations were between 0.827 and 0.990 for the subscale scores and 0.984 for the total score. Inter-tester kappas for each item ranged from 0.200 to 1.000 with a mean of 0.698. Concordance rates were between 50.0 and 100.0% with a mean of 87.2%. Cronbach's alphas for the items of the individual subscales ranged from 0.702 to 0.915. Inter-subscale and subscale-total correlations ranged from 0.503 to 0.820 with a mean of 0.684 and from 0.721 to 0.883, respectively. Cronbach's alpha of the subscales was 0.934. Sensitivity and specificity were 93.3% and 93.8% at the cut-off point of 57.0/57.5 for dementia. Subscale and total scores were significantly different between the demented and non-demented. The total CIDI score was significantly correlated with scores of the Blessed Dementia Rating Scale and the Korean version of the Mini-Mental State Examination.