Facial Expression Recognition (FER) is a among the key elements in the field of human-computer interaction, which facilitates the interpretation of the human being’s emotional state and the response to their emotions by systems. This paper presents a novel 3D approach that leverages 3D triangular meshes combined with Graph Convolutional Network (GCN) to enhance the performance of FER. In this work we propose a method that requires the extraction of 3D mesh data from facial images sourced from the Real-world Affective Database (RAF-DB), which contains 15939 facial images tagged with various expressions. By representing facial geometry as a graph structure, we capture nuanced details that facilitate a deeper understanding of expressions. The processed data is then analyzed GCNs to learn discriminative features essential for characterizing different emotions. We conducted experiments carefully in this study and then observed the high effectiveness of this approach, demonstrated by the impressive recognition accuracy achieved (94.91%). This work not only contributes perfectly to the efforts to advance the state of the art in the domain of FER but also underscores the potential of integrating geometric representations with deep learning techniques for improved performance in emotion detection tasks. Additionally, we discuss the implications of our results and present research expectations regarding the use of 3D techniques and GCN for further advances in this field.