An orthogonal table and a linear multi-regression model were used to optimize the preparation procedure of polyvinylidene fluoride/polymethyl acrilate/cellulose acetate (PVDF/PMMA/CA) blend ultrafiltration membrane, which was prepared via combination of the phase-inversion process and chemical reaction. The pure water flux, the feed solution flux of BSA, the observed retention of BSA, the pore size distribution and SEM images of the membranes were used to estimate the characteristics of the membrane. The results showed that: (1) the combination of orthogonal table and linear multi-regression model is an efficient method to ascertain the optimum preparation conditions for the membrane-making process; (2) solid content, solvent species, additive species, additive content, category of coagulation medium affect characteristics of the membrane, and among them the solid content and additive species are the most remarkable influencing factors; The relationship is Y-4 = 4.921 - 1.871X(1), -0.462X(2) + 0.719X(3) - 0.153X(4) - 0.292X(7), where, Y-4 is a function of simultaneous consideration of the influence of the pure water flux, feed solution flux and observed retention, X-1 is solid content, X-2 is solvent species, X-3 is additive species, X-4 is additive content, X-7 is the interaction between additive content and additive species; (3) when solid content of PVDF/PMMA/CA is 16%, solvent is DMF, additive is PEG2000 with 2%, content of glacial acetic acid (CH3COOH) is 10%, the coagulation medium is an aqueous solution of sodium carbonate (Na2CO3), and gelation time is 30 min, the PVDF/PMMA/CA is a blend of UF membranes with a uniform membrane pore size distribution (0.07-0.10 mu m) and high pure water flux is prepared.