Due to the ineffectiveness of the fixed data window caused by time-varying frequency dynamics and inertia, an online estimation for inertia of power system based on variable data window is proposed. First, the process of power system inertia response is analyzed, and the auto-regressive with extra inputs (ARX) model is used to depict frequency dynamics, on which a parameter identification model is built. Then, based on the adaptive median-mean combined filter (AMMCF), the frequency measurement data are preprocessed. And the fuzzy logic controller (FLC) is used to set the parameters of the forgetting factor recursive least square (FF-RLS) method. Moreover, the indices measuring the identification error and the difference of the results of successive windows are established to realize the online variation of data window length and calculate the system inertia. Finally, case studies are carried out to validate the effectiveness of the proposed approach. The case study results show that the proposed approach can adapt to the changing inertia and is effective to reduce the inertia estimation error.