Membrane recovery of light non-aqueous phase liquids (LNAPLs) was studied. Small membrane modules, operated in a sealed end configuration, were pressurized with several LNAPLs in order that membrane permeability and fouling processes might be characterized. The mitigation of membrane fouling via backpulsing was optimized for backpulse amplitude, frequency, and duration. A full-scale LNAPL recovery device was tested in the laboratory. Clean-membrane permeability was determined via single-fiber extraction of 2-propanol, distilled kerosene, commercial kerosene, commercial diesel fuel, and LNAPL obtained from the subsurface of a contaminated refinery site. Membrane fouling occurred during laboratory extraction of commercial kerosene, distilled kerosene, and the refinery fluid, but not for 2-propanol. Filtration experiments showed suspended solids present within the LNAPLs. Scanning electron microscope images of fouled membranes indicated that these solids were deposited upon the membrane surface. The fouling was due to pore blockage, pore constriction, and/or cake layer formation. The use of backpulsing was found to mitigate pore constriction and cake layer formation. Backpulsing improved recovery rates by similar to30% over a 15 h period, but it could not prevent the occurrence of some irreversible fouling. The device designed to fit into a typical recovery well extracted similar to35% of the volume expected for given operating pressures and clean-membrane permeability, under non-fouling conditions. The less than optimal performance was due to crimping of hollow fibers during the potting process, and not fiber-fiber interactions. Under non-fouling conditions, field operation of the device can yield 1-2000 l per day, which is a theoretical range of subsurface LNAPL recharge rates. With the use of backpulsing under fouling conditions, the membrane recovery device can extract up to 501 per day, but this must be coupled with regular cleaning/replacement of the membranes. (C) 2003 Elsevier B.V. All rights reserved.