Backbone (15N) NMR relaxation is one of the main sources of information on dynamics of disordered proteins. Yet we do not know very well what drives 15N relaxation in such systems, i.e. how different forms of motion contribute to the measurable relaxation rates. To address this problem, we have investigated, both experimentally and via MD simulations, the dynamics of a 26-residue peptide imitating the N-terminal portion of the histone protein H4. One part of the peptide was found to be fully flexible, while the other part features some transient structure (a hairpin stabilized by hydrogen bonds). The following motional modes proved relevant for 15N relaxation. (i) Sub-picosecond librations attenuate relaxation rates according to S2∼0.85-0.90. (ii) Axial peptide-plane fluctuations along a stretch of the peptide chain contribute to relaxation-active dynamics on a fast time scale (from tens to hundreds of picoseconds). (iii) φ/ψ backbone jumps contribute to relaxation-active dynamics on both fast (from tens to hundreds of picoseconds) and slow (from hundreds of picoseconds to a nanosecond) time scales. The major contribution is from PPII ↔ β transitions in the Ramachandran space; in the case of glycine residues, the major contribution is from PPII ↔ (β) ↔ rPPII transitions, where rPPII is the mirror-image (right-handed) version of the PPII geometry, while β geometry plays the role of an intermediate state. (iv) Reorientational motion of certain (sufficiently long-lived) elements of transient structure, i.e. rotational tumbling, contributes to slow relaxation-active dynamics on ca. 1 ns time scale (however, it is difficult to isolate this contribution). In conclusion, recent advances in the area of force field development have made it possible to obtain viable MD models of protein disorder. Following careful validation against the experimental relaxation data, these models can provide a valuable insight into mechanistic origins of spin relaxation in disordered peptides and proteins.