The objective of this study was to evaluate the reliability and validity of a short form of the Severe Impairment Battery (SIB) for a Korean population. Eighty-four Alzheimer's disease patients with Clinical Dementia Rating (CDR) stages 2 (n = 32) or 3 (n = 52) and scores of less than 15 on the Korean version-Mini Mental State Examination (K-MMSE) participated in this study. Seventeen of the patients were men and 67 were women, and the mean age was 75 (SD = 10.4). Cronbach's coefficient alpha of the shortened SIB (SIB-S) was 0.93, and the item-total correlation was significant. Test-retest correlation for the total SIB-S score and subscale scores was significant, with the exception of the 'orienting to name' subscale. Construct validity was confirmed by evaluating the correlation between the SIB-S and the SIB, K-MMSE, CDR, and S-ADL; Spearman correlation coefficients were 0.96, 0.88, -0.67, and -0.63, respectively, which were found to be significant. The difference in the total SIB-S score and its subscale scores between the CDR 2 and CDR 3 groups was significant, except for the 'orienting to name' subscale. More severely demented patients also showed a wide range of the SIB-S score. Finally, the diagnostic accuracy of the SIB-S was high (AUC = 89.5%) in the differentiation of the CDR 2 and 3 patients. The sensitivity and specificity of the SIB-S were 91% and 80%, respectively, when the cut-off score was 32.5. Our results indicate that the SIB-S is a reliable and valid instrument for evaluating patients with severe dementia in the Korean population.