beta-Casein separation by cross-flow filtration was performed after solubilizing beta-casein monomers at low temperature (4-degrees-C) from submicellar aggregates of whole casein in a 1% w/w sodium caseinate solution. Filtration experiments were carried out with 3 different Carbosep membranes, M1 (20 nm), M9 (28 nm) and M6 (80 nm), in a flux-pressure sequence. The maximum purity of beta-casein did not exceed 60% with relatively low permeation flux (< 45 1 h-1 m-2). The linear evolution of the overall hydraulic resistance versus filtered volume and the reversible fouling predominance indicated that permeability was essentially limited by a particle cake build-up on the membrane. The apparent pore diameter of the filtering system, calculated from the experimental transmission of a small size monomer reference solute (gamma1-casein) using Ferry’s size exclusion model, was in the range of 5-6 nm, regardless of the initial membrane pore size. This finding reinforced the hypothesis of a cake fouling, which governed permeability and selectivity. These apparent diameters and theoretical solute sizes allowed a modelled transmission to be calculated for each individual casein and each membrane, which fitted well the experimental results.