This study examined the effect of scoring method, education, and language usage on internal consistency of the Folstein Mini-Mental State Examination (MMSE). Trained bilingual staff administered the MMSE in participants' homes as part of the San Antonio Longitudinal Study of Aging home-based assessment battery. Subjects included 833 community-dwelling Mexican-American (MA) and European-American (EA) elders, aged 65 and older, residing in three socioculturally distinct neighborhoods in San Antonio, Texas. Three methods of scoring the MMSE were examined: serial sevens only, spelling only, and serial sevens or spelling, whichever was higher. Mean MMSE scores±standard deviation ranged from 27.7±2.4 to 28.5±1.9 for EAs, from 25.6±3.2 to 27.2±2.9 for MAs interviewed in English, and from 22.5±4.5 to 25.5±3.5 for MAs interviewed in Spanish, depending on scoring method. Across the three ethnic-language subgroups, the lowest mean scores, largest coefficients of variation, and highest alpha coefficients were observed using serial sevens only. Stratification by educational level showed that alpha coefficients for all three scoring methods were consistently lower in high school graduates than in less-educated groups. Serial sevens only was the only scoring method that yielded acceptably high alpha coefficients across all ethnic, language, and education subgroups. Thus, clinicians should use the serial sevens–only method when administering the MMSE and be alert to the increased potential for false-negatives in more highly educated EA and MA elders, particularly in EAs and MAs proficient in English.