In this thesis, we develop a version of minority game. Minority game as an agent based model has been introduced for analyzing some complicated systems. The model is able to recognize the agents in different environments such as financial markets, road traffic, and transferring the packets through the Internet and variety other of cases. It is a competitive mechanism for determining how to take advantage of limited resources. In the standard form the agents in minority game make decisions using strategy tables. These tables are developed based on the history of the market. In the standard form of minority game, the learning of agents takes place through the process of virtual rating. We equipped the agent of to two layer perceptron neural networks instead of strategy tables. Back propagation algorithm is used as the method of learning of the neural network. Eventually an analogy survey of the agents' performance is carried out between the previous and the new. Results demonstrate that applying two layer perceptron neural networks yields better decisions according to reduced agent attendance and that it enhances performance averages.