In this paper, a sinusoidal extraction method based on a dual observer is proposed. Passing the disturbed harmonic through an special auxiliary filter, measurable variables related to the harmonic frequency can be obtained. Then the unexpected signal is decomposed into a known function and the required estimation term driven by standard signal. The observer is constructed to estimate the required estimation term, thereby the standard sinusoidal signal can be reconstructed directly. This method can represent the disturbance as a parametric form about auxiliary variable, so as to derive the relationship between frequency and harmonic directly, with which the standard signal and disturbance can be decoupled. Because only the unknown parameter driven by standard sinusoidal signal needs to be estimated, the computational complexity is greatly reduced. According to the Lyapunov stability theory, if the bounded noise is absent, the observer can track the standard signal asymptotically. In the multi-disturbances environment, the uniform ultimately bounded performance of estimation error can be guaranteed by adjusting the parameters. Simulation results demonstrate the effectiveness of the proposed dual observation structure.