In order to overcome the difficulty that unknown images can not find best fusion rules adaptively, a new fusion model driven by discrepant features is presented based on dual-color MWIR images. The model firstly finds best image features corresponding with fusion rules on the basis of selecting best fusion evaluation index. Then it uses possibility theory to construct membership function describing the distribution of image features, and establish mapping relationship between discrepant features and fusion rules. Finally, a fusion model driven by discrepant features is established. The experiment results show this model can reflect the true scene better in subjective perception and about 95% of index parameter is superior to others in objective evaluation.