Previous research has shown that some emotions can promote learning while others can hinder it [1]. Emotions are transitory and related to an object. Thus, the measurement of emotions in real time, and in relation to specific learning contents as a learning activity unfolds, is as difficult as it is crucial. Indeed, the relation between emotions and certain specific learning subjects remains to be explored and could contribute to the effectiveness of learning methods, particularly with respect to technologies for learning. The objective of this study is to determine if a relation exists between emotions and learning of physics mechanics during a play session of Mecanika, a Serious Game. Emotions and learning are integrated conceptually through a prediction-action model [2, 3] that shows how bottom-up influences (bodily signals) and top-down influences (performance and conceptual change) relate to emotions. The results of the multilevel analysis show that there is a relationship between affectivity and learning with average affective intensity being higher in participants who demonstrate correct scientific knowledge throughout the game. Contrary to expected results, the analyses indicate that emotions related to high arousal and negative valence do not impair learning or conservation of scientific knowledge in the context of a serious game. These observations align with previous studies showing that negative high arousal emotions do not necessarily interfere with the learning process [4]. This research allowed a sharp look at the moment-by-moment relationship between emotions and learning. The development of online measures of emotions based on multimodal data could be integrated into computer tools for external management of emotions and affective self-regulation contingent on current states of the user to foster learning and performance.