In this paper, we test between suppression and activation accounts of metaphor processing by means of a novel metaphor interference paradigm using mouse-tracking. The goal is to understand how context influences the activation of salient and non-salient features of a concept during the on-line processing of a metaphor. In two mouse-tracking experiments, we examine the activation and availability of conceptual features that were either irrelevant or relevant for understanding a metaphor across various contexts. Our findings support the conclusion that context works primarily by rapidly suppressing salient features of a concept that are not relevant for the particular metaphorical interpretation. What is more, it seems that even further contextual manipulation does not facilitate the activation of non-salient metaphor relevant features.