This paper studies a output-based sinusoidal disturbance estimation method in the presence of multi-disturbances. A low-pass filter is introduced to excite the disturbance characteristics, and series of cascaded unmeasurable signals are then obtained. Particularly, the signals can be decomposed into three parts, including standard sinusoidal virtual signal, equivalent bounded disturbance and decay term. In case that the frequency is known, the auxiliary filter is designed to establish the relationship between frequency and virtual disturbances, with which the external sinusoidal disturbance is reformulated in a parametric form about the virtual disturbance. Resultantly, the input disturbance can be reconstructed through successive order-reduction approach using the estimates of the virtual disturbances. The disturbance observer design method is independent of the controller structure, and removes the necessity to estimate the equivalent disturbance state, which reduces the computational complexity to a large extend. If the equivalent bounded disturbance is absent, the sinusoidal disturbance can be tracked asymptotically with reverse recursive fashion. Otherwise, the uniformly ultimately bounded of estimation error can be guaranteed by adjusting the parameters of the auxiliary filter in presence of multi-disturbances. The effectiveness of the disturbance observer structure is verified by simulation and experiment.