© 2022 IEEE.Tooth caries is one of the most common dental diseases in the world. Although it manifests itself with pain caused by bacterial infection, it is detected as standard in dental radiographs. Different approaches may be required in the treatment of dental caries, the most common of which is dental filling. In this study, a deep learning approach that automatically detects dental fillings on periapical radiographs is presented. RetinaNet, SSD and YOLOv3, which are single-stage detection models, and Faster R-CNN models, which are twostage detection models, were used with ResNet50, ResNet101, VGG16 and DarkNet-53 backbones. Model performances were evaluated with mAP, accuracy, precision, sensitivity, and F1-score. On the other hand, the training times of each model were also recorded. The best performing models were observed as Faster R-CNN ResNet101 and YOLOv3 with 0.92 and 0.99 mAP, 0.87 and 0.8 accuracy, 0.93 and 0.89 F1 score respectively. It was observed that the training process of the models with more layers took longer. These findings show that deep learning models can be beneficial in dentistry, as in many different areas, and there can be decision support units to assist the dentist within the scope of digital dentistry.