Most physical models are approximate. It is therefore important to find out how accurate are the predictions of a given model. This can be done by validating the model, i.e., by comparing its predictions with the experimental data. In some practical situations, it is difficult to directly compare the predictions with the experimental data, since models usually contain (physically meaningful) parameters, and the exact values of these parameters are often not known. One way to overcome this difficulty is to get a statistical distribution of the corresponding parameters. Once we substitute these distributions into a model, we get statistical predictions -- and we can compare the resulting probability distribution with the actual distribution of measurement results. In this approach, we combine all the measurement results, and thus, we are ignoring the information that some of these results correspond to the same values of the parameters -- e.g., they come from measuring the same specimen under different conditions. In this paper, we propose an interval approach that takes into account this important information. This approach is illustrated on the example of a benchmark thermal problem presented at the Sandia Validation Challenge Workshop (Albuquerque, New Mexico, May 2006).