In this paper we present a signal-adaptive algorithm for computing the evolutionary spectra of non-stationary signals. The adaptation is accomplished by estimating the instantaneous frequency of the signal components from an initial evolutionary spectrum. The signal components are obtained applying a masking procedure in the time-frequency plane. Estimating the instantaneous frequency of the signal components permits us to improve the time-frequency resolution of the initial spectrum. The evolutionary spectra is computed using the Gabor representation of the signal. The performance of the proposed signal-adaptive algorithm is illustrated by considering examples of multi-component signals.