This paper summarizes the development of an expert system designed to help solve taste and odor problems in drinking waters which cannot be attributed to typical odorous compounds such as geosmin or 2-methylisoborneol (MIB), but rather to complex mixtures of volatiles usually detected below their individual odor thresholds. Long-term studies at four different sites were carried out to create a data base including flavor profile analysis (FPA) and closed-loop stripping-gas chromatography (CLSA-GC-FID) data. Examination of the data using factorial correspondence analysis allowed us to establish statistical links between odor descriptors and specific volatile organics. A stochastic regression model was also built which permits estimation of a sample flavor profile from its chromatographic content. The strengths, limitations and possible future developments of this method are discussed in detail.