Semantic knowledge is thought to be at least partially grounded in sensory, motor, and affective information, acquired through experiences in our inner and outer world. The reactivation of experience‐related information during meaning access is called simulation. In the affective simulation account, it is assumed that the grounding information depends on the concepts’ concreteness. Whereas abstract concepts are thought to be mainly represented through affective experiential information, concrete words rely more on sensory‐motor experiential information. To test this hypothesis, we measured facial muscle activity as an indicator of affective simulation during visual word recognition. Words varied on the dimensions of concreteness and valence. Behavioral and electromyographic data were analyzed with linear mixed‐effects models with maximal random effect structure to optimize generalization over participants and word samples. Contrary to this hypothesis, we found a valence effect in the m. corrugator supercilii only in response to concrete but not to abstract words. Our data show that affective simulation as measured with facial muscle activity occurs in response to concrete rather than to abstract words. More concrete words are supposed to have higher context availability and richer visual imagery, which might promote affective simulation on the expressive level of facial muscle activity. The results are in line with embodied accounts of semantic representation but speak against its predominant role for representing affective information in abstract concepts.