Food aromas are generally perceived as unitary aromas, i.e. fried potatoes smell like fried potatoes and roasted coffee smells like roasted coffee. Nonetheless, nearly all food aromas are produced by a multitude of volatile components that contribute to an extensive collection of aromas in various compositions. Of the hundreds of volatiles released from fried potatoes and roasted coffee a major se!ection does not produce an odour, many others do produce an odour that contributes to one of both aromas, many others contribute to both. In spite of this, food aromas are not being perceived as collections of discernable odours.Central theme of this thesis is a method used to fractionate, identify and characterise the odorous components in mixtures of odorants: gas chromatography olfactometry (GCO). GCO entails the pressurised transfer of volatiles through a capillary column, after capturing these from the headspace of a foodstuff. Since the column delays volatiles differentially, panellists may sniff these volatiles sequentially on their release from the capillary. This allows the separate sensory evaluation of components in the aroma mixture.A well-established practice in technological studies is that measurement reliability of instruments is estimated and minimised. Although sniffing panels are generally employed as instruments that assess the odour impact of chemical components in food aromas, reliability assessment is generally not applied to their application. Paradoxically, there are no practically available valid methods that may assess the reliability of panel responses. In addition, studies of systematic bias of panel responses in GCO studies are not available, although some common assumptions in GCO studies are not valid psychologically. For instance, intense odorants that arc qualitatively similar to the aroma quality are generally identified as character impact components. However, perception studies showed that odorants may affect (viz. suppress) each others odour contribution in a mixture. Therefore, assumptions regarding the contribution of singular odorants to aromas on basis of GCO, is premature.This thesis introduces methodology to estimate GCO response reliability and reports the empirical testing of this methodology (chapters 2, 3 and 4). In addition, several GCO assumptions considered psychologically disputable, are evaluated in empirical studies. These studies entail the effect of stimulus context on the estimated reliability of GCO results (chapter 4), the evaluation of the contribution of GCO-identified 'character impact components' to an apple model aroma (chapter 5), the effects of task instruction on the identification of odorants in food aromas (chapter 6), the effects of peri-threshold components on a supra-threshold food aroma (chapter 7) and the effects of sequentially presented odorants on their mutual odour intensities as a function of qualitative odour similarity (chapter 8). The general conclusions of this thesis are, that panel responses in the temporary absence of odorants must be used to estimate the reliability of panel odour detections, and that mechanisms that systematically affect response variation should be known and considered m GCO.