The purpose of the present study was to evaluate which statistical model - linear, logarithmic, quadratic or exponential - best described the fatigue-related electromyographic (EMG) changes of back and hip muscles. Twenty healthy volunteers performed a modified Biering-Sorensen test. The EMG activity of the latissimus dorsi (LD), longissimus thoracis pars thoracis (LTT) and lumborum (LTL), iliocostalis lumborum pars thoracis (ILT) and lumborum (ILL), multifidus (MF), gluteus maximus (GM) and biceps femoris (BF) was measured bilaterally using surface electrodes. Higher R(2) values were found for the quadratic models (p<0.05 for all muscles), and lower R(2) values for the logarithmic models (p<0.05 for LTT, LTL, ILL, MF and GM). The exponential models generated higher R(2) values compared to the linear ones for the LTT, LTL and MF (all p<0.05). Further analyses revealed, however, that these models did not add useful additional information, and therefore would only increase the complexity. The findings of the current study validate the use of simple linear regression techniques when studying fatigue-related EMG median frequency characteristics of back and hip muscles during isometric contractions.