Detecting structural changes in time series is an important problem because such structural changes induce abnormal forecast errors. And there are many change types and so, mis-identification of such changes also leads to failures of forecating applications. In this thesis, our main argument is focused on the problem of identification of a transient or level change in exponential smoothing. Our identification procedure consists of two main steps. At first, we estimate the most possible change point and in the next step, calculate posterior probability of occurrence of any change types at the estimated possible change point and detect occurrence of change based on the posterior probabilities. In the estimation of the most possible change point, our estimate is a point which maximize the posterior probability ratio of change to no change at that point with adaptive priors which obtained from past informations and proper initial priors. Performance of the presented method is examined using artificial data and simulations. The results show small mis-identification rate and rapidity in identification of changes. This algorithm can be extended to the case of more change types but initial priors for change types are difficult to derive because they require multivariate form.