So-called mixed metaphors have not received much attention in cognitive linguistic research, despite acknowledgments to the fact that the combination of metaphors is in fact pervasive. This paper makes the case that mixed metaphors present a unique test case for existing theories of metaphor, in particular Conceptual Metaphor Theory and Blending Theory, since these theories make different predictions with regard to the comprehension of mixed metaphors. It will be argued that mixed metaphors selectively combine aspects of semantically conflicting source domains into one figurative meaning. The argument will be made through a two-tiered empirical study that uses quantitative corpus data as well as experimental evidence.