COMATOSE ( complete matrix analysis torsion optimized structur e) is a structure-refinement program based on the quantitative calculation of 2D NOE intensities taking into account the effects of network relaxation and spin diffusion (J. W. Keepers and T. L. James, J. Magn. Reson. 57, 404 (1984)). The macromolecular structure is defined by fixed bondlengths and two-bond angles, leaving only torsion angles and residue orientations as variable parameters. The performance of this structure refinement algorithm with idealized and experimental 2D NOE intensities for oligonucleotides is discussed. The structural information available from COMATOSE is compared with the distances obtained by analysis of the 2D NOE intensities, and with the distances available from the direct solution of the intensity eigenvalue problem. The most commonly used calculation of distances relies on the assumption that the intensities can be approximated by assuming that the individual spin pairs are essentially isolated from all other protons. This approach systematically results in underestimation of distances. Direct solution of the eigenvalue problem when all intensities (i.e., relaxation pathways) are not accounted for results in large errors especially at longer distances (greater than ∼3 A). COMATOSE is capable of optimizing the structure in favorable cases and, at least, yielding a set of reliable proton-proton distances (relative error ∼10% ). In most practical applications, an iterative approach to the development of the structure should probably be taken. An initial trial structure is subjected to COMATOSE to obtain accurate distance estimates from the 2D NOE. These in turn can be used as constraints in energy minimization of the structure to arrive at an improved structure for recycling into COMATOSE.