A reliable site-specific estimate of the individual N-H bond lengths in the protein backbone is the fundamental basis of any relaxation experiment in solution and in the solid-state NMR. The N-H bond length can in principle be influenced by hydrogen bonding, which would result in an increased N-H distance. At the same time, dynamics in the backbone induces a reduction of the experimental dipolar coupling due to motional averaging. We present a 3D dipolar recoupling experiment in which the H-1,N-15 dipolar coupling is reintroduced in the indirect dimension using phase-inverted CP to eliminate effects from rf inhomogeneity. We find no variation of the N-H dipolar coupling as a function of hydrogen bonding. Instead, variations in the H-1,N-15 dipolar coupling seem to be due to dynamics of the protein backbone. This is supported by the observed correlation between the H-N-N dipolar coupling and the amide proton chemical shift. The experiment is demonstrated for a perdeuterated sample of the alpha-spectrin SH3 domain. Perdeuteration is a prerequisite to achieve high accuracy. The average error in the analysis of the H-N dipolar couplings is on the order of +/-370 Hz (+/-0.012 angstrom) and can be as small as 150 Hz, corresponding to a variation of the bond length of +/-0.005 angstrom.