Graphical models are useful to detect multivariate association structures in terms of conditional independencies and to represent these structures in a graph. When fitting graphical models to multivariate data, the uncertainty of a selected graphical model cannot be directly assessed. In this paper, we therefore propose various descriptive measures to assess the uncertainty of a graphical model based on the nonparametric bootstrap. We also introduce a so-called mean graphical model. Simulations and one real data example illustrate the application and interpretation of the newly proposed measures and demonstrate that the mean graphical model performs better than a single selected graphical model.