To explore some mechanisms of generalization in concept formation, we build a three-layer neural network with feedback and Hebbian learning rules. Using binary sequences as input, we simulate the generalization process from multiple examples to a concept. After tens of training, the outputs of the network will converge to stable states which denote the formation of a concept. We suggest that generalization is a nonlinear emergence phenomenon generated by collective behavior of neurons and feedback between neurons is a necessary factor.