An optical coherence tomography (OCT) prediction algorithm is designed and tested on a data set of sample images (taken from vegetables and porcine tissues) to demonstrate proof of concept. Preprocessing and classification of data are fully automated, at a rate of on a standard computer and can be considered to deliver in near real-time. A data set consisting of nine groups was classified correctly in 82% of cases after cross-validation. Sets of fewer groups reach higher rates. The algorithm is able to distinguish groups with strong visual similarity among several groups of varying resemblance. Surface recognition and normalizing to the surface are essential for this approach. The mean divided by the standard deviation is a suitable descriptor for reducing a set of surface normalized A-scans. The method enables grouping of separate A-scans and is therefore straightforward to apply on 3-D data. OCT data can reliably be classified using principal component analysis combined with linear discriminant analysis. It remains to be shown whether this algorithm fails in the clinical setting, where interpatient variation can be greater than the deviations that are investigated as a disease marker.