Understanding the types of defects is of practical interest, which could help developers adopt proper measures in current and future software releases. As the amount of bug reports increasing, manual classification brings a heavy burden to developers. In this paper, we propose a word2vec based framework of multi-granularity automatic classification for bug reports based on fault triggers. Except classifying bug reports into bug/non-bug and Bohrbug/Mandelbug, the classification of Mandelbugs is the focus of this paper. Characteristic representation of common classification methods suffer from data sparsity and high dimensionality, thus we use word2vec, which can express words as low-dimensional word vectors with semantic relations in this paper. Furthermore, in order to improve the quality of classification, we analyzed the impact factors of classification. The results show that our method performs well in automatic classifying bugs into fault trigger classes.