In this laboratory study, nuclear magnetic resonance (NMR) relaxation data were collected on unconsolidated sediment to determine the NMR response, characterized by the mean-log transverse relaxation rate, T2ML−1, and the sum of echoes, SOE, during drainage and imbibition. Measurements were made on four synthetic sands, with a range of grain sizes and Fe content, and two natural loamy sand soils. A porous ceramic plate apparatus was used to induce drainage and imbibition. Water content (θ) was plotted vs. matric potential (ψ) to give the water retention curve (WRC). The drainage and imbibition branches of the WRC were then compared with the corresponding branches of the θ–T2ML−1 and θ–SOE curves. We observed the expected linear trend between NMR signal magnitude and θ. The θ–T2ML−1 or θ–SOE curves did not exhibit drainage–imbibition hysteresis, even though T2ML−1 and SOE varied with θ and the WRC did exhibit hysteresis. Using a simple pore network model, we show that, for well-connected networks, the surface-area/volume ratio of the water-occupied porosity is similar during drainage and imbibition, explaining the lack of hysteresis present in the measured θ–T2ML−1 or θ–SOE curves. For materials with poorly connected pore networks, hysteresis may be observed. We conclude that, for the materials used in this study, it is not possible to distinguish drainage from imbibition using T2ML−1 or SOE. However, because T2ML−1 and SOE depend on θ, NMR data may be useful for characterizing a single branch or an average of the two branches of the WRC by relating these NMR parameters to ψ.