Linear system identification methods combined with neuromechanical modeling enable the quantification of reflex gains from recorded joint angular perturbation, torque, and/or electromyography (EMG). However, the stretch reflex response as recorded by EMG consists of multiple consecutive activation volleys (M1 and M2 responses) separated by a period of reduced activity and is nonlinearly related to joint perturbation. The goal of this study is to assess to what extent linear assumptions hold when quantifying these reflexive responses. Series of ramp-and-hold angular perturbations with fixed velocity but different ramp durations (and, therefore, different amplitudes) were applied to the wrist joint of seven healthy volunteers. Evoked EMG responses were compared to the reflex response estimated from a common linear reflex model relating EMG to perturbation velocity. Model fits described the measured EMG responses best when the perturbation and M1 response durations were equivalent. With increasing perturbation duration, i.e., amplitude, EMG response increased but reflex gain decreased due to the inert period after M1, which is believed to be related to alignment of the refractory period of the motoneurons. For angular joint perturbations exceeding the M1 duration (coinciding with 2° of wrist joint rotation in this study), reflex gain variation may be largely explained from a shortcoming of the linear model in describing the nonlinear reflex response, and in particular the period of low reflexive activity after M1.