The interpretation of stabilization diagrams is a classical task in operational modal analysis, and has the goal to obtain the set of physical modal parameters from estimates at the different model orders of the diagram. The diagrams are contaminated by spurious modes that appear due to the unknown (non-white) ambient excitation and sensor noise, as well as possible over-modelling. Under the premise that spurious modes will vary and physical modes will remain quite constant at different model orders, the focus is to retrieve the physical modes that constitute the identified model, while rejecting the non-physical, spurious modes. Over the last decade, extensive research has been devoted for developing automated strategies facilitating their interpretation. To this end, the interpretation is in principle disconnected from the identification method and boils down to three stages i.e., clearing the diagram from the spurious mode estimates, aggregating the modal parameter estimates in modal alignments and the final parameter choice. Besides the point estimates of the modal parameters, also their confidence bounds are available with some identification methods, such as subspace identification. These uncertainties provide useful information for an automated interpretation of the stabilization diagrams. First, modes with high uncertainty are most likely non-physical modes. Second, the confidence bounds provide a natural threshold for the automated extraction of modal alignments, avoiding the requirement of a deterministic threshold regarding the allowable variation within an alignment. In this paper, a strategy is presented for the automated mode extraction considering their uncertainties, based on clustering a statistical distance measures between the modes. The relevance of the uncertainty consideration in the automated extraction will be demonstrated on vibration data from two bridges.