Laboratory‐scale aquifer cells are commonly used to investigate processes governing contaminant fate and transport in heterogeneous subsurface systems. In recent years, dramatic improvements in image processing methods have led to increasingly refined image resolution in these experiments. With these enhanced imagining methodologies, system parameters, such as nonaqueous phase liquid (NAPL) saturation, can now be quantified at the submillimeter scale. This fine level of resolution, however, is generally inconsistent with the typical scale of a representative elementary volume (REV), the averaging volume associated with property evaluation in continuum‐based flow and transport models. Such inconsistency of scales calls into question the practice of directly comparing laboratory observations with continuum‐based model simulations. This work explores the application of alternative approaches to image data processing for characterization of NAPL source zone architecture in aquifer cells and examines the implications of data processing on the interpretation of experimental‐model comparisons. The utility of two alternative upscaling methods, a continuity‐equation based (CB) and discrete‐block based (DBB) approach are considered. Examination of the stability of point averages of saturation demonstrates that a REV length scale can be defined as ∼30 × d50 for the thickness‐averaged, two‐dimensional aquifer cell experiments examined. Quantification of upscaled source zone metrics, including the average saturation, the second spatial moment in the vertical direction, and the ganglia‐to‐pool (GTP) mass ratio reveals that the GTP is strongly sensitive to observation scale. Use of GTP values computed from upscaled images is shown to improve mathematical model predictions of experimental effluent concentrations by nearly 50%.