Narrow-band imaging (NBI) of surface microvessels of colorectal lesions is useful for differentiating neoplasms from non-neoplasms and for predicting histopathological diagnosis. Furthermore, NBI of surface microstructure, or 'surface pattern', is valuable for predicting histopathology in colorectal cancer. The aim of the present study was to investigate whether surface patterns could be used to predict invasion depth in colorectal cancer, and to compare the accuracy of surface pattern diagnosis in each macroscopic type. Between January 2010 and March 2011, a series of 357 consecutive patients with 378 early colorectal cancers were observed by magnifying NBI and the surface pattern was prospectively evaluated. Surface pattern was classified into 3 types: type I, microstructure was clearly recognised with uniform arrangement and form; type II, microstructure was obscured with heterogeneous arrangement and form; and type III, microstructure was invisible. We also classified the macroscopic type into 3 categories: depressed, protruded and flat elevated. Assuming that type III was an index of massively invasive lesions in the submucosal layer (SMm), the sensitivity, specificity and accuracy were 56.9, 91.7 and 85.7%, respectively. The sensitivity, specificity and accuracy of type III for the diagnosis of SMm in each macroscopic type were: depressed, 88.9, 40.0 and 63.2%, respectively; protruded: 34.8, 96.4 and 90.0%, respectively; and flat elevated, 54.2, 92.7 and 85.0%, respectively. These results suggest that the diagnostic accuracy of surface pattern was insufficient and particularly poor for depressed-type lesions.