This study investigated the use of radon-222 as an in situ partitioning tracer for quantifying nonaqueous phase liquid (NAPL) saturations in the subsurface. Laboratory physical aquifer models (PAMs), field experiments, and numerical simulations were used to investigate radon partitioning in static (no-flow) experiments and in single-well, 'push-pull' tests conducted in non-contaminated and NAPL-contaminated aquifers. Laboratory push-pull tests in a wedge-shaped PAM and field push-pull tests in a NAPL-contaminated aquifer showed that radon was retarded in the presence of NAPL, with retardation manifested in increased dispersion of radon extraction phase breakthrough curves (BTCs). An approximate analytical solution to the governing transport equation and numerical simulations provided estimates of the radon retardation factor (R), which was used to calculate NAPL saturations (S[subscripts n]). Laboratory static and push-pull tests were conducted in a large-scale rectangular PAM before and after NAPL contamination, and after alcohol cosolvent flushing and pump-and-treat remediation. Radon concentrations in static tests were decreased due to partitioning after NAPL contamination and increased after remediation. Push-pull tests showed increased radon retardation after NAPL contamination; radon retardation generally decreased after remediation. Numerical simulations modeling radon as an injected or ex situ partitioning tracer were used to estimate retardation factors and resulted in overestimations of the likely S[subscripts n] in the PAM. Radon partitioning was sensitive to changes in S[subscripts n] in both static and push-pull tests. However, the test results were sensitive to test location, sample size, test design, and heterogeneity in S[subscripts n] distribution. Numerical simulations of hypothetical push-pull tests conducted in a NAPL-contaminated aquifer were used to investigate the influence of homogeneous and heterogeneous S[subscripts n] distributions and initial radon concentrations on radon BTCs and resulting S[subscripts n] calculations. Both of these factors were found to affect radon BTC behavior. A revised method of plotting and interpreting radon BTCs combined with numerical simulations modeling radon as an in situ partitioning tracer (incorporating initial radon concentrations into the model as a function of S[subscripts n]) were used to re-analyze laboratory and field push-pull test BTCs. This method reduced the overestimation of calculated S[subscripts n] values from laboratory tests. / Graduation date: 2003