Backpropagation is a supervised learning algorithm and is commonly used by perceptrons with multiple layers to alter the weights connected to neurons present in the hidden layer. Backpropagation method is widely used as application development to solve the problem. The performance of the algorithm is influenced by several network parameters including the number of neurons in the input layer, the maximum epoh used, the magnitude of the learning rate, and the resulting error (MSE). The performance of the training algorithm is said to be optimum can be seen from the error generated by the network. The smaller the error generated, the more optimal the performance of the algorithm. Tests conducted in the previous study found that the most optimal training algorithm based on the results of the smallest error is the Levenberg-Marquardt training algorithm with an average of MSE 0.001 with level of testing α = 5%. This study analyzes the accuracy of n neurons in layers using the Levenberg-Marquardt training algorithm where (n = 8, 12, 14,16,19). Network parameter which is target error = 0.001, maximum epoch = 1000, and learning rate = 0.01; 0.05; 0.1; 0.2; 0.3; 0.4; 0.5; 0.6; 0.7; 0.8; 0.9; 1. This research uses mixed method which is development research with quantitative and qualitative test (using ANOVA statistic test). The data were collected from random data with 10 input neurons and 1 neuron at the output layer. So from the result of research with 16 neurons in the hidden layer with resulting in the average of the smallest error rate 0.00019584038 ± 0.000239300998 with learning rate = 0.8.