Providing the designers with the relationship between functions and structures is an effective way to accelerate the design process. Currently, the function-structure relationship is usually obtained by theoretically analyzing the transformation from function to structure, such as the FBS model. Though these methods have provided the reasonable explorations of the function-structure relationship, they are complicated and not easy to achieve. Since the function-structure relationship in the existing products also follows the design pattern described by the existing studies, extracting the function-structure relationship from these products instead of the theoretical analysis could also transfer the significant information that what structures could satisfy the required function. Therefore, this paper presents an estimation approach to obtain the probabilistic description of the function-structure relationship in products. First, the product, structure and the functions contained by a product are described with product vector, structure vector and function vector, respectively. Meanwhile, the relationship between them is also defined. Then, a statistical strategy is proposed that treats all the products as the population and the products we gather as the sample, and defines the function-structure relationship as the conditional probability of the appearance of a structure given a function in the gathered products. Afterwards, the maximum likelihood estimation (MLE) is employed to estimate the conditional probability. Compared with other methods, the proposed approach replaces the theoretical analysis with discovering the products, which avoids the complicated modeling and description of the function-structure relationship. In case study, some experiments have been carried out, and a plug-in tool is developed to implement the application of the extracted function-structure relationship in product design. The results have shown the feasibility of the proposed approach and demonstrated the practical value in engineering.