To determine the technical and economic feasibility of cross-flowmicrofiltration on an industrial scale, the expected decline of permeation flux must be predicted taking into account the variability of juice's fouling potential. However, the main difficulty is finding representative parameters. Two semi-empirical models - gel polarisation and mechanistic - were used to fit experimental permeation flows using initial juice turbidity as surrogate for the volumetric concentration of particles in the feed juice. The experimental data of different banana, pineapple, and blackberry juices fittedwell in both models. Although the mechanistic model more accurately estimated the permeation flux density, for practical application, the simpler polarisation model was preferred. Because this method uses a factor (i.e. turbidity) that reflects juice's fouling potential, it allows the optimisation of processing parameters and the prediction of permeation flux range in real industrial conditions.