A multistage affinity cross-flow filtration (mACFF) process for protein purification is proposed. The process is mathematically modelled taking into account a case of rapid equilibrium binding of a target protein to its macroligand. The process performance, i.e., dimensionless breakthrough volume (Qb+)and recovery yield (REC) to obtain a desired purity is analysed by computer simulations. The results indicate that Qb+increases with the increase of stage number (n) due to the increase of affinity binding efficiency. In addition, REC also increases with the increase of n, especially for lower affinity systems, even though the feed loading is the same as the corresponding breakthrough volume that increases with n. Thus both feed loading and recovery yield can be enhanced by raising the stage number. Incompletely permeable membranes reject the target and contaminant proteins. So they delay the appearance of the breakthrough point and compromise the contaminant washing efficiency. Hence although Qb+increases with the increase of membrane rejection coefficient (R), REC decreases when the feed loading equals that of Qb+. However, when the feed loading is kept unchanged and equals Qb+at R=0, REC does not decrease, but slightly increases with the increase of R. This result indicates that incompletely permeable membranes may also be employed for the mACFF process. In general, the model gives a predictive evaluation of the mACFF process successfully.