The invention discloses a protein subcellular interval prediction method using a bag-of-word model. A protein sequence is segmented with a sliding window method, a collection of a large number of sequence words is obtained, sequence word characteristics are obtained through amino acid composition, clustering analysis is carried out on the sequence word characteristics so that a dictionary can be built, then, the bag-of-word characteristics of the protein sequence are obtained through statistical calculation, and finally, the bag-of-word characteristics are sent into a support vector machine multi-class classifier to be processed through protein subcellular interval prediction. It can be proved through experiments that with the protein subcellular interval prediction method using the bag-of-word model, recognition accuracy can be effectively improved, and can be especially and obviously improved in the subcellular classes, where a traditional method is low in prediction accuracy, and the method has important effect on accurately predicting the subcellular positions of unknown protein.