Prostate cancer automatic grading has attracted a lot of attention during the last years . Many research efforts have been fixated on the development of computerized recognition and classification systems to automatically grade Gleason patterns. Automatic computerized Gleason grading methods can be classified into two basic classes: image textural-based class and tissue structural-based (nuclear architecture, gland morphology) class. To the best of our knowledge, tissue structural classification based on three-class classification results including Gleason grade 3, 4 and 5 carcinoma were not reported. The goal of this article is to: (1) develop computerized assessment support systems to automatically grade Gleason patterns 3, 4 and 5 by integrating gland morphology and architectural features; (2) improve classification accuracy especially between intermediate Gleason grades 3 and 4. Computer simulations show an average correct classification accuracy of 97.63%, 96.57% and 87.30% when distinguishing Gleason 3 vs. Gleason 4, Gleason 3 vs. Gleason 5, and Gleason 4 vs. Gleason 5 respectively. These results lead the way towards providing an effective and promising software tool in automatic prostate cancer histological Gleason grading.