This paper proposes an iterative scheme for worst-case parameter estimation in the presence of bounded model uncertainties. The algortihm distinguished itself from other estmation schemes, such as errors-in-variables and Hoo methors, in that it leads to less conservative designs since it explicitly incorporates an a-priori bound on the size of the uncertainties. It also employs an exponential weighting scheme where data in the remote past are given less weight than the most recent measurements. This feature is especially useful in tracking problms where recent observations carry more information about the current value of the unknown parameter. Simulation results are included to demonstrate the performance of the recursive scheme.