Abstract: In this paper we expose new method to analyse time series. The method is essentially applied for the prediction of time series among approximation and modelisation. It is based on wavelet decomposition combined with autoregressive models. An iterative procedure is applied and the performance of the estimator is measured by standardized error. Finally, a comparison with some existing models and methods has been pointed out to prove the performance of our’s. [Copyright &y& Elsevier]