Fuzzy logic is commonly useful to represent the human logic behavior, and in particular in the design of both decision-making systems and classifiers. A big part of its accuracy is provided by the membership functions which usually are selected from a traditional group, e.g., triangular, pi and gamma functions, without considering the data behavior in the scope of application. Therefore, a wrong selection of it may have a negative effect on the accuracy of the decisions and classifications of the Fuzzy logic-based systems. In order to address this issue, in this paper, we propose a method for discovering membership functions, according to the data behavior in the scope of application. The proposal covers two processes, in the first, the guidelines for data preparation are provided, and in the second process, the discovery stages of the membership function are described. According to an evaluation of a model prototype, the proposal enables to confirm whether a traditional membership function is the most suitable, or alternatively, it allows to discover other special functions such as the |sinc(x)|. In relation to the |sinc(x) function, it's concluded that it can be a great choice to emulate the periodicity, which is a feature commonly seen in the data behavior in certain scopes of application.