How the information content of an unfolded polypeptide sequence directs a protein towards a well-formed three-dimensional structure during protein folding remains one of the fundamental questions in structural biology. Unfolded proteins have recently attracted further interest due to their surprising prevalence in the cellular milieu, where they fulfill not only central regulatory functions, but also are implicated in diseases involving protein aggregation. The understanding of both the protein folding transition and these often natively unfolded proteins hinges on a more detailed experimental characterization of the conformations and conformational transitions in the unfolded state. This description is intrinsically very difficult due to the very large size of the conformational space. In principle, solution NMR can monitor unfolded polypeptide conformations and their transitions at atomic resolution. However, traditional NMR parameters such as chemical shifts, J couplings, and nuclear Overhauser enhancements yield only rather limited and often qualitative descriptions. This situation has changed in recent years by the introduction of residual dipolar couplings and paramagnetic relaxation enhancements, which yield a high number of well-defined, quantitative parameters reporting on the averages of local conformations and long-range interactions even under strongly denaturing conditions. This information has been used to obtain plausible all-atom models of the unfolded state at increasing accuracy. Currently, the best working model is the coil model, which derives amino acid specific local conformations from the distribution of amino acid torsion angles in the nonsecondary structure conformations of the protein data bank. Deviations from the predictions of such models can often be interpreted as increased order resulting from long-range contacts within the unfolded ensemble.