Question and Answering Sites are useful for quick and efficient support. In order to recommend questions to proper answerers, it is necessary identify the question subject. In this paper, we presented a comparison among classical information retrieval models (Boolean Model, Vector Space Model and Probabilistic Model) for the question/topic classification task. This study was conducted aiming the development of a Web Service to offer NLP treatment for questions written in Portuguese and this question/topic labeler is one of its functions. The results of our analysis showed that Vector Space Model outperforms the others.