BACKGROUND: The serial 100-7s subtraction, an item on the Mini-Mental State Examination (MMSE), is well known for being difficult for uneducated people. Therefore, we investigated into alternative serial subtractions for serial 100-7s subtraction in uneducated people. METHODS: One hundred sixty-nine subjects were enrolled by neurologic or neuropsychiatric out-patient clinics in 4 university medical centers. The subjects were divided into two groups: an uneducated group and an educated group (at least primary schooling) by questionnaire. We investigated the correlation between incorrect number of serial subtractions and Global Deterioration Scale (GDS) score in both groups and undertook receiver operating characteristic (ROC) curve analysis. MMSE including serial 40-4s subtraction, serial 20-2s subtraction, and serial 10-1s subtraction instead of serial 100-7s subtraction were arbitrally named MMSE4, MMSE2, and MMSE1. RESULTS: In the educated group, serial 100-7s subtraction showed the highest correlation with GDS score (correlation coefficient, 0.465; P < 0.001). In the uneducated group, serial 40-4s subtraction showed the highest correlation with GDS score (correlation coefficient, 0.608; P < 0.001), and serial 100-7s indicated the lowest correlation (correlation coefficient, 0.378; P = 0.023). In ROC curve analysis for MMSE, MMSE4, MMSE2, and MMSE1 to assess the presence of dementia (GDS score > or = 3) in uneducated subjects, the area under the curve (AUC) was 0.648, 0.770, 0.758, and 0.711, respectively, and in educated subjects, AUC for MMSE, MMSE4, MMSE2, and MMSE1 was 0.729, 0.719, 0.716, and 0.714, respectively. CONCLUSION: Out of MMSE items, serial 100-7s is adequate in the educated elderly, but may be less adequate in the uneducated elderly. Serial 40-4s seems to be more appropriate for MMSE in the uneducated elderly.