A new algorithm analyzing neighborhood conditions of adenomatous tissue is is introduced. Using O'Callaghan's definition of neighborhoods, a graph theory approach for measuring histomorphological structures can be created as follows: glands are defined as vertices and the coherence of neighboring glands as edges. The procedure leads to an unoriented, well-defined graph which contains information usually not measurable by conventional morphometric analysis. Measurements on healthy mucosa, tubulo-villous adenoma and highly to moderately differentiated adenocarcinoma of colon revealed statistically significant differences (p less than or equal to 0.05) for the following parameters: number of vertices, number of edges, frequency distribution of n-stars and of n-closed paths. Correct separation and reclassification of 83% of cases could be carried out using discriminant analysis. 11/15 cases (73%) could be classified correctly in a prospective group based upon the learning set. The significance of these findings for automatic pattern recognition in histopathology is discussed.