The feasibility of low-field nuclear magnetic resonance (LF-NMR) was investigated to predict the dielectric properties of Chinese yam slices. Changes in relaxation behaviors and dielectric properties (at 915 and 2450 MHz) of samples during microwave vacuum drying as well as their relationship were studied. Results showed that the total moisture content decreases gradually over drying time, and the free water was removed first, followed by the immobile water and bound water. Correspondingly, the dielectric constant (ε′) gradually decreased till to reach a relative and stable low value; while the loss factor (ε″) changed slightly first, then decreased gradually and also reached a low value finally. Univariate linear models showed that the signal intensity of free water peak (A23) and the transverse relaxation time of immobile water (T22) had good correlation with dielectric properties. Furthermore, partial least squares regression (PLSR) models with four NMR parameters as variables gave better results with R2P of 0.946, 0.936, 0.962 and 0.921 for ε′ and ε″ at 915 MHz and ε′ and ε″ at 2450 MHz, respectively. The overall results revealed that LF-NMR is suitable for predicting the dielectric properties as a rapid and noninvasive method. •Drying caused significant changes in the T2 curves.•Moisture-dependent dielectric properties (DPs) were analyzed.•PCA analysis indicated the relation between NMR parameters and DPs.•PLSR model with 4 NMR parameters showed better prediction accuracy for DPs.