Modern digital music libraries have grown to contain a very large number of musical representation and retrieving images from them may be difficult for people with no prior experience. This study presents a comparison of several convolutional neural networks (CNN) architectures performance on music sheet classification, which are state-of-The-Art computer vision methods to perform classification tasks. The models were trained using randomly selected sheets from different sheet music books and used to classify the source book of the validation data. To evaluate the models with incomplete images, we divide each image of our dataset in nine equal parts, then test the models with them. Performance evaluation of the CNNs prove that they can be very effective in this task.