OBJECTIVE: To determine the effectiveness of novel computerized algorithms adopted from the field of signal processing for better distinguishing between normal tissue, premalignant, and malignant colonic lesions. STUDY DESIGN: The study included 3 groups of colonic biopsies with normal, adenoma, and adenocarcinoma cases. Histological slides were automatically scanned and analyzed with image processing software. Textural variables were obtained from the gray level co-occurrence matrices (entropy, contrast, correlation, homogeneity) and from wavelets analysis. Fractal analysis quantified complexity. RESULTS: The univariate analysis revealed 126 variables to significantly differentiate between normal versus adenoma and adenocarcinoma (p < 0.0001). Multivariate analysis singled out multiple wavelets and co-occurrence matrix-based variables of texture and architectural complexity to be independent predictors of the 3 study groups. Discriminant scores differentiated well between all 3 groups, with sensitivity ranging between 80-99.2% and specificity between 65-99.4%. CONCLUSION: In this novel study we showed for the first time that a combination between digital methods of texture, wavelets, and architectural complexity analysis of the tissue structure was able to automatically and accurately differentiate between normal tissue, adenoma, and adenocarcinoma of the colon in endoscopic biopsies. © Science Printers and Publishers, Inc.