An adaptive exponential smoothing method is proposed by use of the change detection statistic for random level change in exponential smoothing, where the mean, the variance, and the limiting distribution of the statistic are derived. The comparison results with Trigg and Leach's and the exponential smoothing method by simulation show that the proposed method gives the smaller mean square error than the exponential smoothing for almost all cases, while it performs better than Trigg and Leach's for frequently occurred large level changes.