Cone Beam Computed Tomography (CBCT) is a medical imaging technique widely used in dentistry including dental implant planning. To determine the size of the dental implant, it is necessary to detect the alveolar bone at the implant site. In this study, we propose automatic detection of alveolar bone from CBCT images of teeth using the YOLOv3-tiny method. The YOLOv3-tiny network architecture consists of a seven-layer convolution networks and six max-pooling layers in the Darknet-53 network with two output branch scale predictions. CBCT images of teeth obtained from 4 patients consisted of 800 coronal slices of 2D grayscale images, containing 830 alveolar bone annotations. Before the training process, the ground truth image annotation was made in the form of a bounding box on the alveolar bone object. The detection results of the YOLOv3-tiny model were compared with the detection results of the YOLOv3 and YOLOv2-tiny models. The results of the experiment on 640 training images and 160 testing images showed that YOLOv3-tiny outperformed YOLOv2-tiny with mAP of 98.6% and 96.73%, respectively. Meanwhile, shows the same good result as YOLOv3.