Traditional vision-based UAV formation control law have less capability of considering constraints.A new formation control law which can explicitly consider constraints on states and control inputs was designed for vision-based formation flight based on Model Predictive Control (MPC).Control inputs of the follower UAV at each sampling instant were obtained by solving a finite horizon optimization control problem.To estimate the acceleration of the leader UAV an estimation algorithm was proposed based on the difference between predicted states and measured states in the framework of MPC.Simulation results show that the designed formation control law can drive UAVs to achieve the prescribed formation quickly and maintain the configuration in the presence of leader maneuvers.The proposed acceleration estimation algorithm is effective and has low estimation error.