Background: It is a chalange to build a program thatbehave intelligent like human. Specially how to build aprogram that able to learn in order to solve a specificproblem. In this reserarch will be developed a computerprogram implementing an artificial neural network thatlearns to recognize handwritten digits. The focus onhandwritting recognition because it is an excellentprototype problem for learning about neural network ingeneral.Aims: Building an intelligent program that able torecognize hand written number (digits).Methode: Artificial Neural Network withBackpropagation architecture and Stochastic gradientdescent learning algorithm wich is implemented in Pythonprogramming language.Result: The program can recognize handwritten digitswith an accuracy over 96 percent, without humanintervention.Conclusion and Advice: A good learning algorithm thatbe learned with bad learning data will perform worse thana simple learning algorithm that be learned with goodlearning data. It is suggested that this research to beimproved with capability to recognize handwritten letter.