A method for controlling an electric actuator in an exercise device, comprising: supplying a first load set point (FA, kA) upon a displacement of the load element in a first direction, supplying a second load set point (FB, kB) upon a displacement of the load element in a second direction, and detecting an initial position (M) of the moving part of the electric actuator when the reversal of the movement is detected, computing an end-of-transition position (N) exhibiting a deviation in the second direction relative to the initial position, supplying a transition load set point in the form of a monotonic function of the position of the moving part of the electric actuator or of the load element, said monotonic function varying from the first load set point (FA, kA) to the second load set point (FB, kB) between the initial position (M) and the end-of-transition position (N).