This study aimed to verify the relationship between changes in thigh muscle-localized bioelectrical impedance analysis (ML-BIA) parameters and performance in a multiple-set exercise. The sample consisted of 30 female university students (22.1±3.2years). The ML-BIA parameters, including localized muscle resistance (ML-R), reactance (ML-Xc), and phase angle (ML-AngF), were evaluated using a tetrapolar bioelectric impedance device operating at a frequency of 50KHz. The multiple sets protocol was performed with an isokinetic dynamometer. For body composition, total and leg lean soft tissue (LST) were evaluated using dual X-ray absortiometry. Student's t-test for paired samples was used to compare the ML-BIA parameters and thigh circumference pre and postexercise. Linear regression analysis was performed to verify the ∆ML-PhA as a predictor of peak torque for the three sets alone while controlling for total and leg LST. There were differences in the ML-R (∆=0.02±1.45Ω; p=0.001; and E.S=0.19), ML-Xc (∆=2.90±4.12Ω; p=0.043; and E.S=0.36), and thigh circumference (∆=0.82±0.60cm; p<0.001; and E.S=0.16) pre- and post-multiple sets. ΔML-PhA was a predictor of performance in the first set (p=0.002), regardless of total and leg LST. However, the ΔML-PhA lost its explanatory power in the other sets (second and third), and the variables that best explained performance were total and leg LST. The ML-BIA (ML-R and ML-Xc) parameters were sensitive and changed after the multiple sets protocol, and the ΔML-PhA was a predictor of performance in the first set regardless of the total and leg LST.