Methods to estimate step length using accelerometers are gaining attention in recent times. However, the influence of the sensor location on the accuracy of step length estimation is still unknown for models fabricated in a uniform manner. Therefore, the purpose of this study was to compare the accuracy of step length estimations among the following three body parts: ankle, pelvis, and wrist. Ten time-normalized acceleration signals from one gait cycle were obtained from 247 healthy adults aged 20 to 77 while walking barefoot at a comfortable, self-selected speed. Linear multiple regression analyses with leave-one-participant-out cross validation technique were used to build the algorithms. The absolute value of mean error for each participant (AME) was computed to compare the accuracies among the body parts. Mean (standard deviation) values of AME for each part were as follows: ankle, 2.66 (2.24) cm; pelvis, 3.09 (2.39) cm; and wrist, 4.05 (3.01) cm. Statistical analyses revealed significant differences for the ankle-wrist and pelvis-wrist estimations. We found that step length can be estimated from the acceleration signal of the ankle or pelvis with almost the same accuracy (approximately 3 cm of average error between participants). Also, estimation of step lengths with the acceleration signals obtained from the wrist needs to be conducted more carefully than those obtained from the ankle or pelvis.