Here we present a numerical modeling effort to estimate historical groundwater concentrations of perfluorooctanoate (PFOA), perfluorooctane sulfonate (PFOS), and perfluorohexane sulfonate (PFHxS) resulting from releases of aqueous film-forming foam (AFFF) near Colorado Springs, Colorado, U.S.A. AFFF was used during fire-fighting training activities at the Peterson Air Force Base upgradient of the towns of Security, Widefield, and Fountain, CO from 1970 to 2017. All three municipalities’ water supply systems relied heavily on groundwater during this time period. We developed Modflow6 and MT3D-USGS models to help evaluate human exposure to these compounds via drinking water. A Modflow6 model was calibrated to groundwater elevation data within the model domain using PEST, a parameter estimation tool, and three MT3D-USGS transport models were developed for each of the three compounds using the same Modflow6 flow model. Locations and timing of potential source zones were established from Air Force documentation; the masses of contaminants introduced to the water table per training event were estimated using inverse methods to match data collected during a 2018 sampling event of the municipal water wells. These estimates were 2.0 g, 1.0 g, and 3.5 g per training event for PFOS, PFOA, and PFHxS, respectively. The root mean squared error values between observed and simulated concentrations for PFOS, PFOA and PFHxS divided by the range between the highest and lowest observed concentrations were 22.8%, 27.8%, and 34.2%, respectively. Concentrations estimated by our models provided reasonable approximations of historical exposures to PFOS, PFOA, and PFHxS in the Security/Widefield area, but not in Fountain. Our model suggests that the contamination observed in Fountain may have originated from somewhere other than the Peterson Air Force Base. We conducted a Monte Carlo analysis to quantify uncertainty within the model results given available data. The average range between upper and lower 95% confidence limits at our calibration targets was 174 ng/L, 108 ng/L, and 238 ng/L for PFOS, PFOA, and PFHxS, respectively. However, a mass balance between the PFASs assumed to be applied on the surface during training events and the calibrated mass loading to the water table indicates that 99.5% of PFOS, 50% of PFOA, and 88% of PFHxS remain unaccounted for. The unaccounted for masses are assumed to be retained in the vadose zone near the source areas and along the transport pathway. The framework employed here may be suitable for a variety of poly- and perfluoroalkyl substance (PFAS) modeling problems where saturated-zone transport is the primary concern. However, a complete picture of PFAS fate and transport that accounts for all released mass is likely not feasible with traditional control-volume finite-difference simulators like Modflow6/MT3D-USGS; this will hopefully be better achieved when the analytical solutions and PFAS-specific simulators currently under development become more readily available.