Abstract: An improved unfalsified control algorithm based on gradient method was proposed to overcome the defects that the unfalsified control algorithm is easy to fall into the poles and has slower convergence speed. The momentum term, which is a parameter variation of the moment before, was added into the adaptation rule to make the parameter jump out of the local extreme poles and improve the system convergence speed effectively. The algorithm just makes use of the collected data, but not depends on the system model. The simulation results verified the validity of the improved unfalsified control algorithm based on gradient method.