Faced with the challenges of facial expression recognition in the absence of vast labeled datasets, we introduce Maximally Utilized Data Semi-Supervised Facial Expression Recognition algorithm, MUDSS-FER. This algorithm harnesses the Expression Recognition Model (ERM) to refine feature extraction and efficiently leverages all unlabeled data to maximise data utilization. Extensive experiments conducted across diverse datasets have revealed that, remarkably, when using a limited number of labelled samples MUDSS-FER outperforms fully supervised algorithms. This approach elevates classification accuracy and underscores the practicality of semi-supervised learning in facial expression recognition, positioning it as a viable solution for real-world applications where labeled data is scarce.