Understanding the effects of contaminants that can diffuse into low‐permeability (“low‐k”) zones is crucial for effective groundwater remedial decision‐making. Because low‐k zones can serve as low‐level sources of contamination to more transmissive zones over time, an accurate evaluation of the impacts of matrix diffusion at contaminated sites is vital. This study compared numerical groundwater flow and transport simulations using MODFLOW/RT3D at a hypothetical site using three cases, each with increasing discretization of the vertical 10‐m thick domain: (1) a coarse multilayer heterogeneous grid based on one layer for each of four different hydrogeological units, (2) a “low‐resolution” discretization approach where the low‐k units were divided into several sublayers giving the model 10 layers, and (3) a “high‐resolution” numerical model with 199 layers that are a few centimeters thick. When comparing the results of each case, significant differences were observed between the discretizations used, even though all other model input data were identical. The conventional grid models (Cases 1 and 2) appeared to underestimate groundwater plume concentrations by a factor ranging from 1.1 to 36 when compared to the high‐resolution grid model (Case 3), and underestimated predicted cleanup times by more than a factor of 10 for some of the hypothetical sampling points in the modeling domain. These results validate the implication of Chapman et al. (2012), that conventional vertical discretization of numerical groundwater flow and transport models at contaminated sites (with layers that are greater than 1 m thick) can lead to significant errors when compared to more accurate high‐resolution vertical discretization schemes (layers that are centimeters thick).