We study in this paper the combination of different concept detection methods for conceptual indexing. Conceptual indexing shows effective results when large knowledge bases are available. But concept detection is not always accurate and errors limit the performances of con- ceptual indexing. A solution to solve this problem is to combine different concept detection methods. In information retrieval the language model- ing approach shows good results and can be used for concepts indexing. As this framework is easily adaptable we propose some fusion models that extend the language modeling approach to combine different detec- tion methods. In this paper, we present previous works on conceptual indexing and conceptual language modeling. Then we investigate several ways for combining concept detections, both on queries and on docu- ment models. Our experiments, on a standard medical collection, show that our fusion models improve the usual conceptual language model, up to 17% on mean average precision. These results show that mixing con- ceptual detections is an efficient way to reduce the impact of detection errors.