Adaptive notch filters (ANF) are known have non-absolute convergence problems which will lead to precision reduction in long-playing frequency tracking. In this paper, we propose an improved ANF based on Steiglitz-McBride method (SMM) for Coriolis mass flowmeter (CMF) whose frequency, amplitude and phase are time-varying based on the random walk model. An monitor is designed to monitor whether the frequency is estimated rightly or the ANF just filters noise. If the frequency of CMF’s signal is missed, we will modify the parameters and restart the ANF to resume the search for the correct frequency again. The particular algorithm of the improved ANF is also put forward. Simulations have verified the effectiveness of the presented method in tracking CMF’s frequency, which shows superior performance comparing to the primary SMM based ANF.