To describe a membrane separator for ternary mixtures of oil-FAME-MeOH, two models of UNIQUAC and support vector machine (SVM) are developed for the permeate composition and permeate flux, respectively. The UNIQUAC model is used to represent the liquid-liquid phase equilibrium of oil-FAME-MeOH. Its prediction results are consistent with the experimental data measured at the temperatures of 293, 313 and 333 K, respectively. For this two-phase system, experimental results show that the oil-rich phase can be rejected by the ceramic membrane while the methanol-rich phase permeates through the membrane. The permeate composition is mainly determined by the feed bulk concentration, and therefore is consistent with the concentration of the methanol-rich phase predicted by the UNIQUAC model. On the other hand, the permeate flux under various operating conditions, such as feed concentrations, temperatures, inlet flow rates and transmembrane pressures, is modeled by the SVM algorithm. Unlike the general data-based neural network model, SVM is especially valuable in the present crossflow filtration problem of small sample sizes. By combining UNIQUAC and SVM with the experimental ultrafiltration of biodiesel mixtures, the predicted membrane separator performance shows no significant lack of fit, and continuous production of biodiesel with both high purity and maximal productivity is also discussed. (C) 2009 Elsevier Ltd. All rights reserved.