The recent development of three-dimensional NMR spectroscopy has alleviated the problem of overlap of resonances. However, also for the 3D experiments resonance assignment strategies have usually relied upon knowledge about spin systems, combined with information about short (sequential) distances. For doubly (15N/13C)-labelled molecules, a novel assignment strategy has been developed. In this paper we address the possibilities of an assignment strategy for proteins, based solely upon the use of NOE data. For this, the 3D NOE-NOE experiment seems most suitable. Therefore, we have made a theoretical evaluation of double NOE transfer pathways in 28 protein crystal structures. We identify 95 connectivities which are most likely to be observed as cross peaks in a 3D NOE-NOE spectrum of a protein. Given the occurrence of one of these 95 connectivities, we evaluate the chances of occurrence for the others. Analysis of these conditional probabilities allowed the construction of five patterns of related, highly correlated cross peaks which resemble the conventional idea of spin systems to some extent and may provide a basis for assignment and secondary structure analysis from 3D NOE-NOE data alone.