The work described in this paper is focused on the reduction of the impact force exerted by the ground on a falling humanoid robot. It proposes the use of variable stiffness to prevent the motors from damaging when falling is inevitable and the only possible strategy is the protection of the robot using its arms. A fuzzy system is used to estimate the low motor stiffness required to reduce the impact force. The stiffness value is obtained using information about the robot velocity at the moment of impact and the desired impact force. The proposed algorithm is implemented in a virtual Darwin OP humanoid robot. The performance of the proposed algorithm is evaluated through the estimation of the impact force.