Palatable, unhealthy food stimuli can be devalued via Go/No-Go (GNG) training that consistently pairs such stimuli with motor inhibition while healthy stimuli are consistently paired with motor excitation. The reason for this devaluation effect is yet unclear and may result from the formation of motivationally oriented stimulus-action associations, response conflict or inferential learning. While the first two accounts focus mostly on the intrinsically negative motor inhibition component of training, the latter account proposes that training effects critically depend on inferences about the valence of motor responses. The present research aimed to disentangle these effects of motor inhibition and response valence in GNG training by varying task instructions about the valence of go and no-go motor responses. In two studies, chocolate stimuli were always paired with motor inhibition (no-go) while fruit stimuli were consistently paired with motor excitation in one condition, while participants in the other condition received the reversed stimulus-motor response pairings. Task instructions either indicated that no-go responses had negative valence (“do not take”) and that go responses had positive valence (“take”), or identified no-go responses as positively valenced (“keep”) and go as negatively valenced (“throw away”). The results show a strong effect of the evaluative framing of go and no-go responses on chocolate evaluations, but no effect of motor inhibition: Chocolate stimuli were devalued following pairings with a negatively valenced response, regardless of whether this response entailed motor inhibition (no-go) or motor excitation (go). These findings align best with an inferential account of GNG training, indicating that devaluation effects critically depend on inferential processes regarding motor response valence. GNG training procedures may, therefore, be optimized by disambiguating the valence of go and no-go motor responses prior to training.