The investigation was undertaken to establish a relationship between key odorants and perceived flavor attributes of dark chocolate as influenced by cocoa fermentation method, roasting and conching conditions, using multivariate data analysis in an attempt to use one variable to predict the other. Eight of the sixteen flavor attributes used by a trained sensory panel to describe and quantify the intensity of attributes in the samples were significantly different (p < 0.05). Roasting significantly reduced astringency in heap-fermented samples but the reduction in tray-fermented samples was not significant. Unconched samples were rated higher in banana attribute than conched samples. Multivariate data analytical tools, Principal Component Analysis (PCA) and Partial Least Squares (PLS) were used to investigate quantitative descriptive analysis and GC-O data and also to relate the two sets of data. PLS 1 models based on single sensory attributes gave better models than PLS2 models based on all sixteen sensory attributes. Ethyl-3-methylbutanoate (fruity, flowery); 2,5-dimethylpyrazine (popcorn); dihydro-2(3H)-furanone, (sweet); linalool oxide (sweet, flowery); benzaldehyde (earthy, nutty) and 2/3-methylbutanal (cocoa, roasted) modeled fruit attribute. It was also possible to model the attribute astringent from the aroma compounds 5-methyl-2-phenyl-2-hexenal (sweet, roasted cocoa), ethyl-3-methylbutanoate and pentyl acetate (green, cucumber). Since fruit attribute was higher in unconched samples and astringent higher in unroasted samples, it may be possible to use the levels of these important aroma compounds as indicators of the sensory attributes fruit and astringent.