In computational proteomics, the peptide identification via interpreting its tandem mass spectrum is an important issue. The classification of b and y ions in the spectrum plays a vital role for improving the accuracy of most existing algorithms. To solve this problem, a classification method based on frequent pattern mining and decision tree is proposed in this paper. First a dataset is established by use of the identified spectrum in which each datum records the ion positions around an ion with b or y type. The discriminative ion frequent patterns (DIFP) of b and y ions are mined with the dataset. And then a decision tree model organizing these DIFPs is proposed for classifying the b and y ions. Finally, we develop an algorithm for the b and y ions classification called B/Y-classifier. The experimental results demonstrate that an accuracy level of 92% is achieved.