An adaptive Fourier Transform (FT) with an optimal window has been proposed for the time-frequency analysis of nonstationary time series. The adaptive optimal window length has been determined by minimizing MSE. The method allows a good estimation of both frequency and amplitude of the spectrum and can be easily applied to the general case of time-varying signals. The evaluation of the proposed approach has been performed on measured time-varying signals from a suspension bridge model and a steel frame model whose data have the typical non-stationary characteristics. The performance of the method has been evaluated by comparing the analyzed modal parameters to the ones from the classic Short-time Fourier Transform (SFT) technique and the calculated values obtained from the finite element model. The numerical results show that the proposed approach can overcome some of the difficulties encountered in the classic Fourier transform technique and can achieve higher computation accuracy.