The number and intensity of wildfires are increasing worldwide, thereby also raising the risk of smoke contamination of grapevine berries and the development of smoke taint in wine. This study aimed to develop five artificial neural network (ANN) models from berry, must, and wine samples obtained from grapevines with different levels of smoke exposure (i) Control (C), i.e., neither misting nor ... [Show full abstract] smoke exposure; (ii) Control with misting (CM), i.e., in-canopy misting but no smoke exposure; (iii) low-density smoke treatment (LS); (iv) high-density smoke treatment (HS) and (v) a high-density smoke treatment with misting (HSM). Models 1, 2 and 3 were developed using the absorbance values of near-infrared (NIR) berry spectra taken one day after smoke exposure to predict levels of 10 volatile phenols (VP) and 18 glycoconjugates in grapes at either one day after smoke exposure (Model 1: R= 0.98; R2= 0.97; b= 1) or at harvest (Model 2: R= 0.98; R2= 0.97; b= 0.97), as well as six VP and 17 glycoconjugates in the final wine (Model 3: R= 0.98; R2= 0.95; b= 0.99). Models 4 and 5 were developed to predict the levels of six VP and 17 glycoconjugates in wine. Model 4 used must NIR absorbance spectra as inputs (R= 0.99; R2= 0.99; b= 1.00), while Model 5 used wine NIR absorbance spectra (R= 0.99; R2= 0.97; b= 0.97). All five models displayed high accuracies and could be used by grape growers and winemakers to non-destructively assess at near real-time the levels of smoke-related compounds in grapes and/or wine in order to make timely decisions around grape harvest and smoke taint mitigation techniques in the winemaking process.