This study examines the potential of neuronal networks and textural feature extraction for recognising suspicious regions in endoscopy under variable perceptual conditions and systematic or random noise in the data. Second-order statistics and discrete wavelet transform-based methodologies are examined in terms of their discrimination abilities, and several neuronal network learning algorithms are compared in terms of success. The results provide numerical evidence that neuronal networks are capable of classifying offline and online tissue samples extracted from standard images and VHS videotape recordings of colonoscopy procedures with satisfactory success rates. This type of technology could prove to be useful for developing intelligent adaptive systems that will assist medical experts in real-time to automate minimally invasive diagnostic procedures.