A new direct design method was used based on closed loop inputoutput measured data for fault estimationand controllers were designed without setting up a state space model for the fault and system.An inputoutput relationship of the closed loop system was used to parameterize a least square problem for fault estimation.When imposing the lower and upper bounds constrain conditions on the faultsthe fast gradient method was applied to solve it.On basis of the fault estimationsa concept of the subspace predictive control was introduced to consider how to design the controller when the control inputs have the affine forms.The mutually iterative steepest descent method was used to obtain some affine terms and gain termsaccording to the separable principle.Finally the effectiveness of the method was demonstrated by an example of helicopter in hovering state.