End‐member mixing models, such as the Keeling and Miller‐Tans plots, are frequently used to interpret isotopic data collected from environments where mass transfer occurs due to diffusive processes, however, researchers do not commonly consider the effect of diffusive kinetic fractionation on assumptions of linearity in these mixing models. Risk and Kellman (2008) recently showed the potential for non‐linearity in the Keeling plot approach, but their simplified model offers only a first order approximation of this effect in complex systems such as soils. Here we use 3‐D numerical simulations and measurements of soil δ ¹³C‐CO₂ flux accumulating in a static head space chamber to conclusively show that in diffusive environments Keeling plots are non‐linear, violating key assumptions of the technique and potentially creating a large source of error in data analysis and interpretation.