We propose a force estimation method in fatigue condition using a muscle-twitch model and surface electromyography (sEMG). The twitch model, which is an estimate of force by a single spike, was obtained from sEMG features and measured forces. Nine healthy subjects performed isometric index finger abduction until exhaustion for a series of dynamic contractions (0–20% MVC) to characterize the twitch model and static contractions (50% MVC) to induce muscle fatigue. Muscle fatigue was identified based on the changes of twitch model; the twitch peak decreased and the contraction time increased as muscle fatigue developed. Force estimation performance in non-fatigue and fatigue conditions was evaluated and its results were compared with that of a conventional method using the mean absolute value (MAV). In non-fatigue conditions, the performance of the proposed method (0.90 ± 0.05) and the MAV method (0.88 ± 0.06) were comparable. In fatigue conditions, the performance was significantly improved for the proposed method (0.87 ± 0.05) compared with the MAV (0.78 ± 0.09).