One of the essential things in research engaged in the field of Computer Vision is image classification, wherein previous studies models were used to classify an image. Javanese Letters, in this case, is a basis of a sentence that uses the Javanese language. The problem is that Javanese sentences are often found in Yogyakarta, especially the use of name tourist attractions, making it difficult for tourists to translate these Javanese sentences. Therefore, in this study, we try to create a Javanese character classification model hoping that this model will later be used as a basis for developing research into the next stage. One of the most popular methods lately for dealing with image classification problems is to use Deep Learning techniques, namely using the Convolutional Neural Network (CNN) method using the KERAS framework. The simplicity of the training model and dataset used in this work brings the advantage of computation weight and time. The model has an accuracy of 86.68% using 1000 datasets and conducted for 50 epochs based on the results. The average inference time with the same specification mentioned above is 0.57 seconds, and again the fast inference time is because of the simplicity of the model and dataset toolbar. This model's advantages with fast and light computation time bring the possibility to use this model on devices with limited computation resources such as mobile devices, familiar web server interface, and internet-of-things devices.