The article discusses approaches to solving the function approximation problem within parametric synthesis of design solutions using the theory of uncertainty. The uncertainty theory allows solving optimization problems under epistemic un-certainty with low computational costs. In many cases, to apply the theory of un-certainty at the stage of preliminary design, we have to solve the problem of ap-proximating experimental data for obtaining analytical expressions that reflect the relationship between parameters of technical objects. In this regard, the choosing an approximation method problem arises that provides high accuracy and the possibility of constructing numerical characteristics of functions that depend on uncertain parameters. We consider the experimental data for solving two arising in practice task, using multiple non-linear regression techniques, cubic splines and neural networks. As a result of experiments had identified the need for pre-processing of the input data to produce satisfactory results. Research results were used to compare methods and identifying advantages and disadvantages.