In the paper we study the efficiency of semantic concept association in multimedia semantic concept detection. We present an approach to automatically learn from the corpus the association strength between pair-wise semantic concepts. We discuss two usages of association strength: 1) applying positive concepts with high association strength for selecting expressive component in the model-based fusion and 2) applying negative concepts with low association strength as filters. We evaluate its efficiency on the task of semantic concept detection on the large-scale news video dataset from TRECVID 2005 development set. Our experimental results demonstrate that exploiting positive association reduces the size of feature dimension in the modelbased fusion and significantly improves the rank performance of system. The mean average precision is increased to 0. 215 on the validation set and 0.206 on the evaluation set. Compared to the traditional model-based fusion, the improvement is about 9.1...