A series of 23 unsaturated bench scale column experiments was carried out to study grain scale processes controlling the efficiency of soil vapor extraction for removal of volatile organic compounds (VOCs) from a desert soil sample (mass fraction of organic carbon, approximately 0.001). Experiments consisted of passing VOC-containing, humidified (>95% relative humidity) air through an unsaturated soil column until breakthrough occurred and then passing VOC-free air through the column until the VOC removal was complete. Effluent VOC concentrations were measured at frequent intervals. Experimental variables included VOC (benzene and p-xylene), soil moisture content (0 to 18% by volume), concentration of VOC in the inlet air stream, and interstitial velocity (0.2 to 0.6 cm s-1). Experimental breakthrough curves were modeled with an advection-dispersion model coupled with a first-order rate equation to describe mass transfer between phases. At 18% moisture, slow water-to-air mass transfer of benzene as the rate-limiting process explains the initial removal of benzene (C/C0>0.1), but the desorption profiles strongly suggest that a slower process (intraparticle pore diffusion) becomes the dominant process at longer times. The breakthrough curves for p-xylene at 10 and 18% moisture also suggest that two sequential processes control removal of p-xylene. The time constant for the faster process is comparable to that determined for the initial rate-limiting process for benzene at 18% moisture. The rate for the second process is the same order of magnitude for p-xylene at 10 and 18% and benzene at 18%. Calculations of the mass distributions of the VOCs among air, water, and soil strongly suggest that the more hydrophobic p-xylene tends to accumulate at the air-water interface (up to 60% of the total mass) and that benzene primarily accumulates in the aqueous phase. ¿ American Geophysical Union 1995