The multicomponent signal with nonlinear timefrequency (TF) struct u re occupies a wide frequency band, and its spectral contents usually overlap. Ai med at the difficulty of extracting the needed component from such complex signa ls, a novel method was proposed to extract components based on instantaneous par ameters estimation. Firstly, FMmlet transform was adopted to determine TF dist ri bution with high TF resolution and less crossterm interference. Then, accordin g to the corresponding relation of energy ridges in TF plane and the IFs, the IF of needed component was estimated with the TF peakdetection method. And the in s tantaneous amplitude of needed component was estimated using least squares algor ithm. Finally, combining the instantaneous frequency and amplitude, the needed c omponent was reconstructed by using the signal model. A synthetic signal was ana lytically formulated and used as a tested signal for evaluating the method perfo rmance. Theoretical predictions and numerical experiment show the feasibility of the method which has good application prospects for indepth feature extracti on of nonstationary multicomponent signals.