Recently, a quantitative J correlation technique has been presented which makes use of homonuclear Hartmann-Hahn cross-polarization (TOCSY) to measure (3)J(C)'(C)' in proteins isotopically enriched with (13)C [Grzesiek, S. and Bax, A. (1997) J. Biomol. NMR, 9, 207-211]. Since homonuclear Hartmann-Hahn is twice as fast as conventional COSY transfer, this method is much less sensitive to transverse relaxation, which is the principal limiting factor in achieving long-range J-coupling correlations in macromolecules. Here we describe a similar experiment which is used to measure(3) J(NN) coupling constants between sequential amide(15) N nuclei in the backbone of ubiquitin. As expected from the low magnetic moment of (15)N, the (3)J(NN) coupling constants are exceedingly small, with values between 0.14 and 0.36 Hz for residues in β-conformations and values below 0.15 Hz for residues in α-conformations. In contrast to what is expected from a Karplus-type dependence on the backbone angle ψ, large differences in the values of(3) J(NN) are observed for a number of residues with very similar backbone ψ angles. A quantitative description of statistical and systematic errors, in particular of relaxation effects during the TOCSY transfer, shows that these differences are highly significant.