The validity of the BMI and fat mass index (FMI) as indicators of obesity was evaluated in a group of 3–5 yr old (n=486) children. Bioelectrical impedance analysis (BIA) was measured (using 50 kHz and tetrapolar electrodes) in order to calculate percent fat mass (%FM) and FMI (fat mass/stature squared). For boys, obesity was defined as ≧20%FM. For girls, the cutoff for obesity was ≧25%FM. However, obesity was defined as a BMI at or above the 90th percentile of age- and sex-specific data in this study. The percentile cutoffs for FMI were the same as for BMI using the same sample. There were correlations between BMI or FMI and %FM, but there was no significant correlation between BMI or FMI and stature. Therefore, it appears that both the BMI and FMI in this study are far more useful indices with which to assess obesity, and are reasonable indicators of fatness. However, with the use of %FM by BIA as the criterion for obesity, BMI and FMI had high specificities (95.5–96.4% for BMI and 99.5–100% for FMI) and lower but variable sensitivities (30.4–37.5% for BMI and 42.9–68.8% for FMI). Thus, almost all children who were not obese were classified correctly. In contrast, many obese children were not correctly identified by BMI and FMI. Therefore, we conclude that BMI should be used with caution as an indicator of childhood obesity. The new recommendations based on the FMI approach for defining childhood obesity are associated with a level of sensitivity that is somewhat higher than that of the BMI approach. Caution should, however, be used in generalizing from the findings in this study, and a further investigation of the issue is required.