Lev-Ari This research aims to develop efficient model-based techniques for non-stationary signal analysis and spectral estimation, with a particular emphasis on Burg's maximum-entropy approach and the order-recursive realization of linear predictors. The ultimate objectives are: 1. to construct numerically-robust adaptive algorithms for estimating the parameters of such models; 2. to establish the statistical properties (e.g., rate of convergence, tracking capability and steady-state error) of such algorithms when applied to both persistent and transient non- stationary signals; and 3. to obtain simple procedures for recovering spectral characterizations, such as the Wigner-Ville distribution or the cyclic spectrum from the estimated model parameters. Adaptive algorithms that efficiently estimate periodically- varying model parameters associated with cyclostationary signals have been constructed. The ultimate goal is to extend these results to a broader family of signals, including stationary, cyclostationary, harmonizable, and asymptotically-mean-stationary signals. In particular, multiresolution wavelet basis functions are used to characterize the dynamics of time-varying model parameters, leading to the construction of adaptive estimation algorithms that are recursive in time, order and resolution.