We used response surface methodology (RSM) to optimize the preparation conditions that had great effects on the performance of the polydimethylsiloxane (PDMS)/ceramic composite membranes for pervaporation. Good performance of membranes could be realized through manipulating three variables, which were polymer concentration, crosslink agent concentration, dip-coating time. In our study, we established the regression equations between the preparation variables and the performance of the composite membranes. We investigated main effects, quadratic effects and interactions of the three variables on the flux and the selectivity of composite membranes. The results showed that polymer concentration was the most significant variable that influenced the permeation flux and the selectivity among three variables and the experimental results were in good agreement with those predicted by the proposed regression models. At a feed temperature of 333 K under a pressure of 500 Pa in an ethanol concentration of 4.2 wt.%, the maximum flux of the 12.95 kg m(-2) h(-1) was obtained by employing the model under the following preparation conditions: polymer concentration 7.4 wt.%, crosslink agent concentration 10.6 wt.%, dip-coating time 60 s. One can expect to apply the regression equations in the preparation of PDMS/ceramic membranes and reasonably predict and optimize the performance of the composite membranes. (C) 2007 Elsevier B.V. All rights reserved.