Upper bones are strong and flexible tissue made up of collagen and calcium phosphate. They mainly contribute to the movement of the human body and serve as a protective shield for the body's soft organs such as the brain, lungs, and the heart. Without these bones, the human body would not be constructed to function ordinarily. However occasionally, due to accidents, an individual is exposed to some diseases such as injury or infection that lead to defects in the regular shape and growth of bone construction. This deficiency in the bone structure is so-called bone abnormalities. Frequently, the preliminary diagnosis of bone abnormalities is made by specialists using X-rays of the patient's injury site to show the shape and density of the bones. They are classified into normal or abnormal. The detection and classification of bones depend on the experience and human effort. So the error in the results of this process can expose the patient to a great danger and catastrophe of his life. Therefore, deep learning algorithms from artificial intelligence were applied to help specialists avoid wrong or inaccurate diagnoses when detecting bone abnormalities in X-ray images by using a pre-trained convolutional neural network called Xception model. The model was customized to fit the bone abnormalities classification then applied to a dataset consisting of 42000 X-rays of the upper bones of some patients collected from Kaggle depository. We trained, validated, and tested the customized Xception model. The proposed Xception model attained Precision (85.20%), Recall (85.13%) and F1-Score (85.07%).