A novel method for the tracking of oscillatory components in EEG signals using adaptive filters is proposed. The tracking capabilities of this adaptive algorithm are achieved by minimizing an oscillation
criterion instead of maximizing the output spectrum power. Therefore it is able to track an oscillation, and describe its evolution over time, even during low-amplitude segments. This provides additional information compared to traditional time-frequency analysis. The proposed method also yields the temporal traces of oscillatory components. This offers an interesting alternative to classical processing where the frequency bands are fixed. This adaptive tracking algorithm can be extended to track simultaneously multiple oscillations, acting as an adaptive filter bank. It can also be generalized to multivariate data analysis, permitting the joint processing of several EEG sensors. The proposed method can be useful in several circumstances. For instance, it is applicable to frequency estimation over time and phase extraction from EEG signals. Indeed, application of our adaptive tracking algorithm, which yields filtered narrowband signals, alleviates the problem of physically interpreting the instantaneous phase of broadband signals. Promising results, in a visual object recognition framework, for frequency estimates in the Theta band and phase-phase couplings (measured with the phase locking value) between the Theta band and the Gamma band are presented and discussed.