In the process of oil sludge treatment, water and oil contents are of particular interest. In this study, the performance of a rapid, accurate analysis of water and oil in oil sludge through low-field H-1 NMR relaxometry, followed by using partial least-squares (PLS) regression models, is reported. In total, 13 oil sludge samples collected from the Hangzhou petroleum refinery are analyzed. Calibration models are developed by PLS regression with full cross-validation on the data obtained using azeotropic distillation by the Dean-Stark method as a reference. The results indicate that calibration is satisfactory (R-cal(2) > 0.99) for both water and oil, when using the raw magnetization decay data for PLS regression; the root mean squared errors of cross-validation (RMSECV) are 0.78% and 0.82% for water and oil, respectively. This shows that low-field H-1 NMR relaxometry is a rapid, reliable, nondestructive, and solvent free alternative for determining the water and oil contents of oil sludge.