Correct approximation of experimental data with significant nonlinearity and heteroscedasticity is relevant. The article presents various approaches to the approximation of experimental data using the ordinary least squares method (LSM) and weighted LSM using medians. A distinctive feature of the data under consideration is the presence of several values in each section. This makes it possible to determine the median for each section. It is shown that the experimental data under consideration are curvilinear, and for their approximation it is advisable to use a quadric. The paper proposes a median approximation, the robustness of which to outliers was confirmed by introducing data contamination. It is shown that medians remain stable in the presence of anomalous points, which makes the method robust. It was also shown that the data are characterized by heteroscedasticity, therefore, to improve the accuracy of the approximation, a weighted least squares method was used with two approaches to calculating the weighting coefficients. Median approximation based on weighted least squares method is considered as a promising approach to improve the robustness of models in the presence of anomalous data and heteroscedasticity.