In view of shortcomings of some methods for similarity measurement, like value dependent of series elements and insufficient mining of information in series, a new method for time series compartmentation, approximation representation and similar measurement is proposed in this paper. Based on sufficient mining of information and orderliness in series, the time series are divided into many sections and the curve fitting model of each section is established. Then, the time series are represented approximately with a sequence of the curvatures of each time in the sections, while the curvature distance is proposed. Finally, the similarity searching algorithms in time series based on curvature distance is proposed. It mines the information of the series sufficiently, retains and recognizes the major shape of the series effectively, experimental results prove the effectiveness, stability and accuracy of the method proposed in this paper.