Efficient trend extraction methods can provide early warnings, severity assessments of monitored subjects and information for decision support. The traditional algorithms for trend analysis of curves include Sliding Window algorithm (SW) and Extrapolation for On-line Segmentation of Data algorithm (OSD), which use total least squares for curve fitting. Compared with conventional least squares, the total least squares has a higher accuracy of fitting a straight line. In addition, since there is no restriction on the maximum length of the sliding window for SW algorithm, the length of window can be very long when threshold for Detection of point becomes larger. As OSD algorithm restricts the minimum length of sliding window, mutations within minimum sliding window cannot be detected for defects of the SW algorithm and the OSD algorithm. This paper presents a new method for trend analysis of data streams. The method uses total least squares to improve the accuracy of trend analysis. It also presents variable sliding window algorithm to solve the fixed window problem with the SW algorithm and OSD algorithm to achieve a reasonable segmentation for data streams. The experimental results show that the method is effective.