In this paper, we proposed an estimation method of the ranges of HRV feature values based on improved decision tree, which used to describe stress. This method extracted HRV features from short-time (2min) ECG signals that obtained from stress-stimulate experiments. Then, multiple decision trees were summarized as a tree with universal recognition performance by utilizing the white-box classification characteristic of the decision tree. Finally, the ranges of the HRV features were extracted from the tree. The method proposed in this paper can display the ranges of the HRV features and had been experimented with 250 ECG signals collected from 114 subjects. The results show, two different states of relaxation and high-stress can be recognized by utilizing the ranges of the HRV features, and the recognition accuracy rate was 86.7%, not less than the traditional classification models. Moreover, the recognition process is simple and the practical application value is high.