Response surface analysis was applied in optimising volatile organic compound (VOC) production by shaken-flask cultures of a selected Candida oleophila strain. A fractional factorial experiment (2(n-k)) was designed by setting four parameters at two levels: pH (3-7), concentration of yeast extract (YE = 5-15 g l(-1)), glucose (C = 10-30 g l(-1)) and (NH4)(2)HPO4 (N = 5-15 g l(-1)). VOCs produced were sampled by means of headspace solid-phase microextraction (SPME), separated and identified by gas chromatography-mass spectroscopy (GC-MS). On the basis of the results of factorial design, YE and C were optimised by performing a central composite design. Second order polynomial models and response diagrams were obtained for total VOCs (VOC I), esters (EST) and isoamyl acetate (IAA) production. A range of variation of YE and C able to maximise simultaneously all three responses has been found. Average final values of responses (VOC I = 125.3 mg l(-1); EST = I S. I mg l(-1); IAA = 16.80 mg l(-1)) were in good agreement with calculated values, thus confirming the reliability of our empirical model in describing VOC production. (C) 2003 Elsevier Inc. All rights reserved.