In a digitized society, biometric systems are gaining increasing importance and are being applied in various fields of application. Facial recognition technology has established itself as one of the most prominent methods for personal identification due to its non-invasiveness and ease of application. Despite significant advances in this area, the reliable identification of children, especially in forensic scenarios, remains a challenging task. Existing facial recognition systems, which are mainly designed for adult recognition, reveal weaknesses in identifying children. This master’s thesis therefore explores the potentials and challenges of retraining facial recognition models for the specific recognition of children through the use of synthetic datasets. The aim is to improve the performance of these systems and to examine their adaptability to the unique characteristics of children’s faces. The results show a significant increase in recognition accuracy after fine-tuning existing models, thereby highlighting the potential of synthetic datasets in biometric research. This work thus contributes to increasing the efficiency and reliability of facial recognition in children and offers important impulses for the further development of this technology in security-critical and forensic contexts.