Backpropagation is a supervised learning algorithm and is usually used by perceptrons with many layers to change weights that are connected to neurons in hidden layers. Pattern recognition is a branch of data mining that focuses on pattern recognition and regularity in data, in some cases it is considered almost identical to machine learning. This study uses 12 training algorithms that aim to determine the accuracy of data pattern recognition for backpropagation network training algorithms in Artificial Neural Networks using 10-14-1 and 10-18-1 neuronal models and statistical inference testing and descriptive statistical analysis with alpha values ( α) = 5%. The conclusion of the study on the 10-14-1 neuron model obtained a significant value of 0,000. Levenberg Marquardt (trainlm) training algorithm has the smallest delta average value that is equal to 0.0047110 which is found in lr = 0.2 and gets the percentage of accuracy of data pattern recognition by 100%. Based on descriptive statistical analysis in neuron models 10-18-1 the Powell-Beale Restarts (traincgb) and Levenberg-Marquardt (trainlm) training algorithms have a percentage in the accuracy of data pattern recognition of 99%. Meanwhile in the inference statistical test obtained a significant value of 0.000 In this model the Levenberg Marquardt (trainlm) training algorithm has the smallest delta average value of 0.0067820 in lr = 0.5.