Computer vision based systems address the need for fast, reliable and non-destructive methods for food quality assessment which is traditionally done by manual inspection techniques that are costly, time-consuming, and high labor intensive. A recent advancement in these systems is the use of hyperspectral imaging (HSI) which can exploit reflectance or transmittance characteristics in a wide range with narrow bands, to achieve improved classification for accurate quality control. In this study, a HSI based computer vision system based on reflectance characteristics is proposed for assessment of dried figs which are economically important for rural development and yet prone to mold infection. By extracting the features as the average intensity of the fig regions. at each spectral band, the proposed HSI system employs sequential floating forward selection with commonly used classifiers (support vector machines and Bayes classifiers), to precisely find contaminated dried figs for their pneumatic removal from the production line. The proposed HSI system can achieve an accuracy of 99.3% based on the most discriminative twenty-seven spectral bands. It also produces an acceptable accuracy of 93% by using only four bands. A preliminary lab-based prototype can control the figs and then pneumatically remove the detected contaminated ones. The throughput of this prototype is seven figs per minute when the full spectrum of 784 bands is evaluated by a single processing line. When only the most discriminative four bands are considered, the throughput improves to seven figs per second using four processing lines on the conveyor belt, which makes it promising for an operational detection system. In addition to reflectance mode, it is shown by an experimental setting that the transmittance characteristics can help identify dried figs with internal contamination when the contamination have no detectable affect on the outer fig surface. (C) 2016 Elsevier B.V. All rights reserved.