Online customized product involves complex processes, including design and production, and hence the quick feedback of the design and its feasibility is important and difficult for implementing online customization. Therefore, a case-based reasoning (CBR) customized method is proposed, which can shorten the time for designing and avoid unnecessary repeated design. Firstly, the weights of demand and engineering design are determined by quality function deployment(QFD) and analytic hierarchy process(AHP). And a more objective, conveniently-handled and easily-distinguished method based on the Gaussian member function is constructed to calculate the matching value. Besides, a machine learning method is constructed on the basis of Beta distribution. Matching threshold can be obtained through machine learning, which changes depending on the market. Secondly, according to customers' diversified sensitivities towards various products, regulatory factors are introduced to adjust the final matching value. By comparing the matching threshold, the design and its feasibility can be acquired. Finally, customized refrigerator is taken as an example to prove the practicability and effectiveness of the method.