Portable NMR devices can measure water content, lipid content, and to show differences between layers of materials and air, using Car-Purcell-Meiboom-Gill (CPMG) sequences [1-4]. These differences have mainly been used for non-medical applications and have never been shown in the single-sided portable NMR device. In this work, we investigate the use of a magnet with flat field isosurfaces over 4 cm (designed by Prado, [4]) for assessment of proton density fat fraction (PDFF). We experimentally demonstrate a correlation between PDFF and apparent T2 in phantoms. The long-term goal is to use this device for in-vivo clinical applications, such as measurement of intra-hepatic and intra-muscular fat. Experiments were performed using a single-sided Neodymium Iron Boron magnet (12x12cm2 top; 7cm height) and a flat spiral RF coil (75 mm diameter) [5]. The magnet’s field strength varied from 0.4T at 0cm and 0.08T at 4cm from the surface. A CPMG sequence measured free induction decays (FID) with parameters: 16.73 MHz center frequency, NSA=100, 2000 echoes, transmit/receive BW= 200kHz/10 MHz, 25 samples/echo, TR=500 ms. Apparent T2 values were estimated by performing a mono-exponential fit to the echo amplitudes. Signal measurements were made 1 mm above the magnet surface. Measurements were acquired in room-temperature mixtures of skim milk and heavy whipping cream with 0, 25, 50, 75 and 100% cream. The measurements were made in succession and the protocol was repeated 12 times. A second experiment measured mixtures of milk and cream with 0, 12.5, 25, 37.5, 50, 62.5, 75, 87.5, and 100% cream with the same protocol but were placed on ice between measurements. PDFF was validated with an IDEAL-SPGR pulse sequence on a 3T HDxt GE MRI magnet. A third experiment measured apparent T2 values of a phantom made from a linear gradient of 3% agar in 100% coconut oil. Two triangular prisms of agar and coconut oil stacked to form a rectangular prism (0.5 cm x 4.5 cm). Apparent T2 measurements of partially-volumed voxels were acquired at 6 points along the agar/oil gradient, 2 points in 100% agar, and 4 points in 100% coconut oil by moving the sample over the coil. Reference MRI scans determined the PDFF of skim milk and cream to be 0.5% and 17% respectively. Figure 1 shows that estimated T2 varies inversely with PDFF. Apparent T2 measurement variability is low enough to produce invertible curves of apparent T2 vs. PDFF with ±0.5% sensitivity for a 0 – 17% PDFF range at 0ºC and ± 2% sensitivity for a 0 – 12% PDFF range at 23ºC. Linear regressions demonstrated R-squared values of 0.93 at 23ºC and 0.98 at 0ºC.We expect the slope of this curve to change with different echo spacing due to diffusion effects [5]; T2 estimates will decrease as echo spacing increases. Figure 2 shows the results of the agar and coconut oil gradient experiment. The left-most measurements are made over pure agar, while the right-most measurements are made over pure coconut oil. A partial voluming effect can be seen where measured apparent T2 value is the approximate average of T2 measured in pure agar and pure coconut oil. The data demonstrates that we are sensitive to the entire dynamic range of fat. The non-linear relationship between apparent T2 and PDFF is likely related to the shape of measurement zone.The milk-cream experiments show an increased negative slope of apparent T2 vs. PDFF at low temperatures. This documented temperature dependence is a likely result of crystallization of fat in milk [6]. Apparent T2 can be measured regardless of temperature; temperature and diffusion coefficient must be known to estimate PDFF. The results show the single-sided device is capable of demonstrating a previously derived relationship between T2 and temperature in fat/water emulsions. Because body temperature is stable under healthy conditions, this effect will not confound future in-vivo studies. The agar/coconut oil gradient experiments show that apparentT2 can estimate the fat/water ratio in a partially-volumed voxel. The non-linearity of the T2-PDFF relationship remains a challenge. We expect this to be resolved in applications where the measurement zone is much larger than fat/water aggregates. In cases where the scale cannot be avoided, tuning of the pulse bandwidth may help overcome error.We demonstrate that a single-sided NMR device can measure relative PDFF in a time-efficient manner and with sensitivity of ±0.5% at 0ºC and ±2% at 23ºC. This magnet geometry, once scaled to an appropriate size for humans, may be useful for the assessment of intra-hepatic or intra-muscular fat for detection of diseases such as fatty liver disease or metabolic syndrome. Calibration at body temperature and decoupling of diffusion effects is pending.