Segmentation of time series is one of the important tasks in time series data mining. Segmentation has two major uses: It may be performed either to detect when the system that creates the time series has changed or to create a high level representation of the time series for indexing, clustering, and classification. Approaches to on-line segmentation of time series are necessary when identifying and predicting temporal patterns in real-time time series databases are needed, and this is the focus of this paper. A formal description of segmenting time series problem and a criterion for the evolution of segmentation algorithms are presented. An on-line iterative algorithm of segmenting time series, called OLS (on-line segmentation), is then proposed. OLS is independent of a priori knowledge about the segmented time series. Experimental results demonstrate that OLS can on-line detect the critical change points of time series with less 憃ver fit?than that of competitive algorithms.