Frequency estimation is one of the fundamental issue in systems theory and signal processing, and has applications in different areas of electrical engineering such as control and system identification, biomedical engineering, control and power system protection, communication and radar. In the recent decade, researchers have focused on an algorithm of frequency estimation of sinusoidal signals based on notch filtering (NF). This algorithem which is called multi adaptive NF (ANF) has been composed of N subfilters in parallel, each subfilter estimate one of the component of a quasi periodic signal including K sinusoid signals. For this structure, three types of stability in cases can be investigated. In the case of , frequencies of quasi periodic signal are isolated equilibrium points of multi ANF and exponential stability of these equilibrium points are investigated and proved. In the case of , multi ANF has a continuum of equilibrium points. In this case, the concept of semistability has been used and proved that the estimated frequencies of K subfilter converge to frequencies of quasi periodic signal and the estimated frequency of N-K subfilters change slowly and converge to the nearest input signal frequencies after a relatively long time. In the case of , estimated frequencies have bias and variance and ultimate boundedness of estimated value under periodic disturbance is investigated. Especially, when N=1 and the sinusoidal signal is accompanied by undesirable signals, has more practical significance. These undesirable signals can be harmonic or nonharmonic components of the sinusoidal signal, white and colored noise. In this thesis, besides introducing multi ANF and studing its characteristics, the previously mentioned cases for multi ANF are investigated. To improve ANF performance in the presence of undesirable signals, approachs have been offered. The most important approach offered in this thesis is using window function in the frequency estimation loop of ANF. The background of using window function goes back to the discrete Fourier transform (DFT) and its modified forms, which improves this algorithm in signal processing. The investigations in this thesis shows that window function has favorable effects on estimation of sinusoidal signal parameters by ANF in the presence of undesirable components. Keywords: Frequency estimation, Adaptive notch filter, Semistability, Ultimate boundedness, Harmonics, Window function