Atmospheric dispersion models are usually off-line coupled to mesoscale meteorological models. This may generate the loss of mass consistency, defined as the conservation of uniform mixing ratio. We investigate in this short paper the impact of the resulting mass consistency errors. Three methods based on the renormalization of density, on fluxes computed with mass mixing ratio and on the adjustment of the vertical velocity, respectively, are aimed at reducing the mass consistency errors. They are benchmarked and applied to two test cases: air quality modeling over Europe for summer 2001 (typical of reactive dispersion) and simulation of the Chernobyl accident (typical of passive dispersion). Our tests indicate the differences between the passive and the reactive cases. The investigation of the spatial patterns (especially of the vertical distribution) discriminates the method based on the adjustment of the vertical velocity. Indeed, this method suffers from the enhancement of the numerical diffusion (illustrated in the passive case) and from the modification of the escape flux (for ozone).