Abstract: Traditional methods for stock index research are still at the stage of judging by experience or relying on simple data analysis, among which fundamental analysis and trading indicator analysis frequently used. These methods have noticeable disadvantages: Inefficient utilization of existing information or requirement for highly experienced users. A modified fuzzy time series (FTS) model was proposed based on the following three aspects. Firstly, a new method of interval division was developed. Secondly, a new weight formula for fuzzy set was devised. Thirdly, a modified FTS model was built with the application of SVM classification model. Predictions for stock index were made by using the proposed model. Experiment results from Shanghai index data ranging from 1996 to 2003 indicate that compared with other important FTS models; the proposed model provides better performance.