Internet traffic exhibits a rich autocorrelation behavior, responsible for curving the Energy/Averaging function. We show that the traffic exhibits variations of its details in many different time scales (multiresolution structure), which can account for this feature. We relate the curving of the Averaging function to such "interesting" time scales, show that it is possible to "read" these time scales (levels) directly off the Averaging function, and propose some methods to accomplice that, also applicable to stochastic processes emerging in other fields (e.g. finance).