An extension of the approach adopted in Part I was developed to improve the identification of diffusivity (D) in relation to the inverse mass diffusion problem. It was based on an estimation of mass fluxes from concentration profiles previously fitted with smoothing spline functions. Data was smoothed to remove noise and to compensate for low spatial resolution. Estimated D values were compared with the true solution used to generate virtual concentration data. The error (e(r)) on D was considerably reduced if the Fo domain was not uniformly distributed but it verified uniform distribution of the average concentration. The effect of the quality of the experimental concentration data (spatial and temporal resolution, noise intensity) on e(r) was also studied. Where D varied with concentration, e(r) was minimal and close to the noise intensity if the numbers of concentration profiles or concentration data per profile were higher than 20. (c) 2006 Elsevier Ltd. All rights reserved.