The partitioning of total electron density into localized molecular orbitals (LMOs) within the source function framework is presented as an exploit useful to the characterization of noncovalent interactions. The resulting approach uses only a converged LMO base as input and does not require grid data handling or numerical integration, making it viable for the study of large systems. Tests conducted on a series of prototypical hydrogen and halogen bonds demonstrate that this descriptor retrieves chemically intuitive interaction motifs beyond the classical two-center picture. Advantages and disadvantages of different localization schemes are discussed in terms of their peculiar description of sigma-pi mixing and in terms of their inherent basis set dependence.