A cut test is done on fermented and dried beans as the most commonly used method to visually assess bean quality for trading and before it is further processed. It provides information on the internal colour, any signs of diseases and pests and the degree of fissuring. This information is used to select the optimum roasting conditions to process into liquor and sensory evaluations of cocoa liquor and chocolate. The steps involved in cutting the beans for analysis are easily implemented, however the visual assessment requires experience to yield reliable results, and a significant amount of time to process. The objective of this research is to develop a tool by applying deep learning models and to read images of these cut tests to classify them quickly and reliably. This first step involved selecting a variety of cut test images from the extensive library of over 1,000 cocoa beans samples from more than 55 origins of producers participating in the Cocoa of Excellence Programme. A neural network model for semantic segmentation to identify the cut beans within these images was created. The next step involved using cut test images of recently submitted bean samples, along with a number of other bean samples of lower quality to broaden the scope of the attributes used in the network training. From these images, each cut bean was isolated and labelled by an experienced evaluator for each property, providing over 20,000 labelled sub-images. The segmentation model was then applied along with the labelled data to train models each on colour, fissuring, and defects. The segmentation model works well for cut test images, with over 95% test accuracy. The deep learning models for colour and fissuring on these segmented cut bean sub-images currently demonstrate respective prediction accuracies of 78% and 73% tested against the ground truth classifications identified by the experienced evaluator. The accuracy for defect detection is to be determined, as more images of low quality and defective cut beans are required. The results of this research demonstrate a proof of concept for providing a digital tool to evaluate and report on cut test images of fermented and dried cocoa beans. Such a tool would increase capacity of cocoa producers to get immediate feedback on the characteristics of their beans. Keywords: Cut tests; Cocoa bean quality; Deep learning; Physical quality; Automated testing