<p>The tail-sitter unmanned aerial vehicles are utilized extensively nowadays since they merge advantages of both rotary-wing UAVs and fixed-wing UAVs. However, the attitude control of tail-sitter UAVs suffers from unknown nonlinearities and disturbances due to the wide flight envelope. To solve the problems, the compensation function observer-based additive state decomposition (CFO-based ASD) controller is proposed. According to the principle of ASD, nonlinear terms are put to the secondary system. Both non-matching and matching disturbances are incorporated into the primary system, where cascade control is implemented. In order to improve the robustness, the disturbance terms are separately estimated using the CFO. Additionally, the UAV is modeled as two parts to avoid the problems of stalling due to high angle of attack. For the problems of over-reliance on aerodynamic parameters in the traditional modeling approach, the aerodynamic moments are obtained by cross product of force and force arm. The numerical simulations of Matlab/Simulink and hardware-in-the-loop simulations of Rflysim are implemented, which are used to compare the model compensation control and PID simulations. The results confirm the superiority of the CFO-based ASD controller in disturbances rejection.</p>