The Stefan column consists of liquid A evaporating into an inert/stagnant gas B with a sweeping B stream at the top. It was designed to estimate binary gas diffusivities, D-AB's, but "end effects" such as gas mixing at the top and interfacial curvature have been either ignored or uncorrelated to the operational settings. This study's hypothesis is that gas mixing at the top and the gas-phase aspect ratio affect D-AB estimation in the acetone (A)-ambient air (B) system at 50 degrees C. The sweeping stream Reynolds number (Re) and the gas-phase aspect ratio (AR = initial gas phase height to column internal diameter) were the variables tested. Isothermal evaporation-diffusion experiments were conducted in which the temporal interfacial descent was tracked. The settings were 492 <= Re <= 5378 and AR between 5 and 15. A 1D transport model allowed determination of the experimental diffusivity, D-AB,D-exp, by nonlinear regression. For Re < 600, the D-AB,D-exp errors relative to D-AB,D-CE (predicted by the Chapman-Enskog kinetic theory for low-density gases) were small and unrelated to AR, while for Re > 600 the errors increased considerably with Re and were inversely proportional to AR. This study is the first to relate the column's operational settings to the D-AB estimation errors. The column should be operated at low sweeping gas Re and large AR for accurate D-AB,D-exp's. The low Re region deserves further study, while the present transport model may have to be replaced by computational fluid dynamics simulations to account for the multidimensional gas flow patterns.