The objective of this study was to develop a calibration model between time-domain low-field nuclear magnetic resonance (LF-NMR) measurements and dry matter (DM) content in single potatoes. An extensive sampling procedure was used to collect 210 potatoes from eight cultivars with a wide range in DM content, ranging from 16 to 28%. The exponential NMR relaxation curves were resolved into four mono-exponential components using a number of solution diagnostics. Partial least-squares (PLS) regression between NMR parameters (relaxation time constants T(2,1-4) and magnitudes M(0,1-4)) and DM content resulted in a model with low error (RMSECV, 0.71; RMSEP, 0.60) and high correlation (r(CV), 0.97; r(test), 0.98) between predicted and actual DM content. Correlation between DM content and each of the proton populations revealed that M(0,1) (T(2,1), 3.6 ms; SD, 0.3 ms; r, 0.95) and M(0,4) (T(2,4), 508 ms; SD, 53 ms; r, -0.90) were the major contributors to the PLS regression model.