(3) (3)J(H)N,(H alpha), (3)J(H)N,(C beta), and (3)J(H)N,(C') couplings, all related to the backbone torsion angle phi, were measured for the third immunoglobulin binding domain of protein G, or GB3. Measurements were carried out using both previously published methods and novel sequences based on the multiple-quantum principle, which limit attenuation of experimental couplings caused by finite lifetimes of the spin states of passive spins. High reproducibility between the multiple-quantum and conventional approaches confirms the accuracy of the measurements. With few exceptions, close agreement between (3)J(H)N,(H alpha), (3)J(H)N,(C beta), and (3)J(H)N,(C') and values predicted by their respective Karplus equations is observed. For the three types of couplings, up to 20% better agreement is obtained when fitting the experimental couplings to a dynamic ensemble NMR structure, which has a phi angle root-mean-square spread of 9 +/- 4 degrees and was previously calculated on the basis of a very extensive set of residual dipolar couplings, than for any single static NMR structure. Fits of (3)J couplings to a 1.1-A X-ray structure, with hydrogens added in idealized positions, are 40-90% worse. Approximately half of the improvement when fitting to the NMR structures relates to the amide proton deviating from its idealized, in-peptide-plane position, indicating that the positioning of hydrogens relative to the backbone atoms is one of the factors limiting the accuracy at which the backbone torsion angle phi can be extracted from (3)J couplings. Introducing an additional, residue-specific variable for the amplitude of phi angle fluctuations does not yield a statistically significant improvement when fitting to a set of dynamic Karplus curves, pointing to a homogeneous behavior of these amplitudes.