Nanocrystalline silicon oxide (nc-SiO ₂ ) immunosensor array-based electronic tongue (E-tongue) has been recently reported to simultaneously detect multiple food toxins with subfemtomolar sensitivity. However, the quantification in these reports is quite imprecise leading to an error of more than 100%. In this paper, the quantification accuracy of multiple food toxin detection in the subfemtomolar range has been improved by more than 90% through upgraded design of the E-tongue system by incorporating two major modifications. First, the pore geometry of the nc-SiO2 immunosensors has been optimized to obtain the best combination of sensitivity, selectivity, and reproducibility through the evaluation of a figure of merit. Second, in the multivariate data processing using partial least squares discriminate analysis, additional input parameters corresponding to selectivity and standard deviations of the experimentally measured data have been incorporated. The final set of input parameters include peak frequency corresponding to maximum impedance sensitivity, bandwidth of the impedance sensitivity characteristics, cutoff frequency from noise spectroscopy, and their standard deviations. The optimized E-tongue system is capable of quantifying 0.1 fg/ml Aflatoxin B1 and Ochratoxin A with an error of only 10% and 20%, respectively, which is a remarkable achievement in the domain of food toxin detection. The proposed E-tongue system is low cost with minimal operator dependence and hence has immense potential for commercial deployment.