Small-scale spatial variability of chemical concentrations in soil samples can introduce substantial uncertainty in the calculation of average concentrations and chemical mass in non-aqueous phase liquid (NAPL) source zones. Centimeter-scale sampling and analysis of a 2.5-m-long soil core from an experimental release of dense NAPL into a sandy aquifer at Canadian Forces Base Borden, Ontario, Canada, yielded chemical concentrations that varied by a factor of 1,000 times over a vertical distance of less than 0.5 m. Simulations to assess the effect of meter-scale sampling more typical of routine site investigations illustrated that there will be substantial uncertainty in average chemical concentrations calculated from less frequent sampling. For soil sampling at a 0.5-m interval, there would be only a 55 percent likelihood that the calculated average perchloroethylene (PCE) concentration would be within ±50 percent of the average calculated by detailed sampling. For sampling at a 1.25-m interval, there would be only a 30 percent likelihood that the calculated average PCE concentration would be within ±50 percent of the average calculated by detailed sampling. This degree of uncertainty is substantial in the context of designing an in situ treatment remedy or evaluating the performance of a mass removal remedy. Based on these results and the spatial variability in chemical concentrations anticipated in most NAPL source zones, the soil sampling and analysis performed for most routine site investigations will be of limited value in making useful estimates of contaminant mass.