An empirical correlation between the peptide 15N chemical shift, delta 15Ni, and the backbone torsion angles phi i, psi i-1 is reported. By using two-dimensional shielding surfaces delta (phi i, psi i-1), it is possible in many cases to make reasonably accurate predictions of 15N chemical shifts for a given structure. On average, the rms error between experiment and prediction is about 3.5 ppm. Results for threonine, valine and isoleucine are worse (approximately 4.8 ppm), due presumably to chi 1-distribution/gamma-gauche effects. The rms errors for the other amino acids are approximately 3 ppm, for a typical maximal chemical shift range of approximately 15-20 ppm. Thus, there is a significant correlation between 15N chemical shift and secondary structure.