Recent theories of cognition have proposed that language comprehension is achieved by mentally recreating, i.e., simulating, the sensory and motor information present when actually experiencing a described event. In order to understand the meaning of a single word, we simulate the experience we had with its referent, in case of nouns, or action it denotes, in case of verbs. To test this theoretical proposal we conducted an experiment with a visual lexical decision task. We preselected 126 nouns whose referents are typically associated with making a movement toward or away from the body (e.g., a cup or a lance) and 108 verbs that imply actions toward or away from the body (e.g., to pull or to throw). We then asked participants (N1 = 117) in an online norming study to rate on a five-point Likert scale how typically these words are associated with movements toward or away from the body. We calculated the correlation between estimates of opposite implied directions and excluded words that were outliers. The final set of stimuli consisted of 115 nouns (correlation between opposite implied directions: r = -.95, p =.000) and 90 verbs (r = -.92, p =.000). These words were used in the experiment in which participants (N2 = 22) had to decide whether the presented string of letters was a word of the Serbian language or not, using a custom built response box that induced movements toward or away from a participant's body. From the embodiment theory, prediction was that if a word recognition requires sensorimotor simulation, congruence between the direction of a response movement and that of word's simulated action must be facilitatory, while incongruence should lead to slower responses. Nevertheless, a mixed-effects modeling showed that after controlling for the random effects of participants and stimuli, as well as the effects of control covariates (word length, word frequency, order of presentation and word type /noun vs. verb/), there was no significant interaction between action direction implied by the word and the direction of movement required for the response. However, the measure of word valence was moderately correlated (r =.51, N = 195, p =.000) with measure of implied direction and, in addition, showed barely significant effect on response latencies (b = -0.01, SE = 0.004, t = -1.96, p =.040): the more pleasant the noun/verb was, the faster it was recognized as a word. Previous studies showing effects in support of embodied language made use of stimuli with clear sensorimotor content. In this research we tried to move one step forward by investigating what we thought of as a whole distribution of action words. However, our findings are not in line with previous studies, which might mean that the embodied nature of majority of representations is hard, if not impossible, to capture on a behavioral level.