The paper looks at the structure of fuzzy rule-based models from the point of view of a function relating membership grades of inputs with rule outputs. This view in turn is generalized by an approach that produces the output functions of the fuzzy rules using input and output data. In this view, a formulation to compute the output of the model consists of estimating the parameters of the output functions. Essentially, the paper suggests an alternative method for fuzzy modeling based on output functions constructed from level sets and input and output data. The data driven method provides an easy and efficient way to develop and process fuzzy models. Examples of function estimation problems show that the data driven level set method is very effective when compared with alternative modeling techniques.