Real-time frequency information is extremely important in many fields of mechanical engineering, such as fault diagnosis, noise and vibration control, underwater acoustic detection, vehicle communication, etc. However, sometimes frequencies cannot be directly detected, making it important to quickly and accurately estimate the frequencies from contaminated signals of a mechanical system. An adaptive notch filter (ANF) is one of the most popular methods for online frequency estimation due to its simple structure and low computational complexity. However, ANF is a biased estimation if the signal contains uncorrelated noise. An enhanced adaptive notch filtering (EANF) method, which is able to reduce the frequency estimation bias and improve the estimation speed from contaminated signals, is proposed in this paper. Firstly, the limitations of the traditional ANF method are theoretically and numerically analyzed. Then, the principles of the proposed EANF method are formulated, including key parameter optimization and uncorrelated noise compensation in the update process. Afterwards, a multiple extension of the proposed EANF method is constructed using the adaptive simultaneous structure. The results of the numerical simulation show that the proposed method is superior to traditional ones. Finally, a two-stage vibration isolation system is established for experimental validation. The experimental results also demonstrate the effectiveness of the proposed method.