Backpropagation is one of the methods contained in a neural network that is able to train dynamic networks using mathematical knowledge based on architectural models that have been developed in detail and systematically. Backpropagation itself is able to accommodate a lot of information that serves as a useful experience. However, the Backpropagation Algorithm tends to be slow to achieve convergence in obtaining optimum accuracy and requires large training data and the optimization used is less efficient. The purpose of this research is to optimize the learning rate on backpropagation neural networks. Source of data obtained from CV. Bona Tani Hatonduhan. There are 3 network architecture models used in this study, namely 2-51, 2-6-1, and 2-7-1 with learning rates of 0.1, 0.2, and 0.3. the results of trials carried out with MATLAB software produced the best architectural model, namely the 2-7-1 model with a learning rate of 0.3 with an accuracy of 83%. Based on this background, it is hoped that the results of the research can help in the process by optimizing the learning rate of the backpropagation Neural Network on the selection of the best architecture.