Backpropagation is a supervised learning algorithm and is usually used by perceptron with many layers to change the weights connected to neurons in the hidden layer. These algorithms need to be tested to get the most appropriate training algorithm in data pattern recognition. In this research, 12 backpropagation network training algorithms were tested, namely traincgf, traincgp, traincgb, trainscg, traingd, traingda, traingdm, traingdx, trainrp, trainbfg, trainoss, and trainlm using two models, 15-20-1 and 15-25 -1. The parameters used are target error = 0.001, maximum epoch = 10000, input neurons = 15, output neuron = 1, neurons in hidden layers = 20 and 25, 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. Based on the results of statistical tests using inference statistical analysis with alpha (α) = 5%, showed that trainlm was the most optimal algorithm on the model of 15-20-1 and 15-25-1 with the average delta of 0.00577 and 0.00515. While using descriptive statistical analysis the results obtained that the algorithm traincgb, traincgf, traincgp, trainlm, trainrp, and trainscg are optimal algorithms on models 15-20-1 and 15-25-1 with the percentage accuracy of data pattern recognition of 100%. From inference and descriptive testing, information was obtained that the trainlm algorithm is the most optimal algorithm for 15-20-1 and 15-25-1 neuron models in the accuracy of data pattern recognition, so that this algorithm can be used as a basis for developing applications in the field of artificial neural networks.