In this work, a computational framework for detecting and analyzing the continuous students’ emotions in a real-time online classroom was proposed. A video conferencing system suitable for the online classroom was developed using Web Real-Time Communication (WebRTC). Following this, a Convolutional Neural Network (CNN) model was proposed to predict the emotional valence and arousal of the students. The AffectNet dataset containing labeled emotions of 291 651 images were used to develop and test the proposed model. The predicted valence and arousal values were used to determine students’ emotional states. Five emotional states—neutral, bored, confused, hopeful, and curious—suitable for online classroom teaching were selected. It was found that our proposed CNN-based model outperforms the current baseline model.