Introduction. Membrane filtration is widely applied for separation and concentration of micro- and nanocolloids. For example, cross-flow microfiltration of milk is used to separate casein micelles (retentate) from serum proteins (filtrate) and to concentrate the micelles in the retentate up to the volume concentration factor VCF = 1…3.5. The process is efficient, when the permeate flux Jp, the solute (serum proteins) transmission TR, and the final VCF are high. However, generally Jp and TR decrease with increasing VCF. To overcome this problem, concentration-separation is carried out in a continuously operating system having N successive stages: filtration modules used at the stage i (i = 1…N) operate at a fixed VFCi and the retentate produced at the stage i – 1 is used as a feed for filtration modules of the stage i, so that VCF1 < … < VCFi–1 < VCFi < … < VCFN. Each stage i can operate at different transmembrane pressures TMP and retentate cross-flow rates v (that is rarely considered in the industrial process optimization). The goal of our work is to design the multi-stage separation-concentration system in: 1) minimizing the membrane surface for a system operating at desired number of stages N and final retentate (casein micelles) concentration VCFN; 2) maximizing the solute (serum proteins) concentration and recovery yield in the permeate. The approaches developed so far to design membrane filtration system, rarely account for complex relation between the process efficiency and the operating parameters. Experimental/methodology. A general theoretical approach was developed to describe the solute recovery yield in the permeates obtained and the membrane surface required at different stages of continuously operating multi-stage filtration system. The development is based on the mass balance (theoretical), which is combined with empirical data of Jp and TR at various operating conditions (VCF, TMP, and v). The approach allows determining optimal values of VCFi, TMPi and vi , i.e. determining for every stage the set of operating conditions required to obtain the highest total yield of serum proteins in the filtrate with the use of the minimal total membrane surface. The approach was illustrated with the case of milk proteins separation by microfiltration of milk. Pilot scale experiments were carried out in a batch mode at 50°C using the UTP (uniform transmembrane pressure) system equipped with 0.1 µm ceramic membrane under different operating parameters: v = 6.0, 6.5, and 7.0 (m/s), TMP = 0.4, 0.7 and 1.0 (bar). In every experiment VCF was varied in a step-wise mode from 1.0 to 3.5 and dependencies Jp(VCF) and TR(VCF) were obtained and used for the theoretical multi-stage membrane system design. Results and discussion. A design method based on theoretical mass balance approach and experimental relationships between Jp and TR as a function of operating parameters was proposed. Besides of the theoretical approach development and application, particular attention was paid to the fact that for 0 < TR < 1 concentrations of transmittable solute (e.g. serum proteins) in retentate with given VCF differ between filtration systems operating in batch (usually lab and pilot) and continuous multi-stage (usually industrial) modes. As soon as in a general case Jp and TR can depend on concentration of any component of the retentate, this calls into question the applicability of the experimental data typically obtained in a batch system for description of continuously operating multi-stage systems. However, specially designed experiments on filtration of milk enriched with serum proteins demonstrated that this effect can be neglected during the design of milk filtration under the studied operating parameters for VCF < 3.0. The approach can then be used to design milk microfiltration system.