This research establishes a “Customized modular product system”. Modular products can be defined according to their different functions and components. To build the learning and searching evaluation model of this system, the back-propagation neural networks and gray theory were used. Besides, Fuzzy theory and analytic hierarchy process were also utilized to model the relation and evaluation algorithms. In this way, it can provide neural networks with basic learning rules and a training samples. Based on different degrees of the consumer demand, the proposed method in this research performs the optimal module-combination for modular products. It will guide the customers to select the most suitable modular product group. Additionally, through the study case of the baby stroller manufacturing company, how to achieve the purpose of customization can be illustrated.