Abstarct: The purpose of this thesis is offline positioning of a moving body on the ground using the Earth's magnetic field components. For this purpose,by using WMM2015 Earth's magnetic field model, data of Earth's magnetic field is generated , then established a multilxayer perceptron neural network and by using changing process of Earth's magnetic field and adjust the parameters of neural network by using the metheod of supervised learning back propagation error and by using two different steepest sescent and hybrid learning algorithms the positioning system has been constructed. simulation results show that the speed of convergence network adjustable parameters to optimal values in the hybrid learning algorithm is more than steepest sescent algorithm. positioning system established by using multilxayer perceptron have suitable accuracy and robustness and can be positioned with an accuracy of about 100 meter average.