Despite being the main oil palm (Elaeis guineensis Jacq.) producer in the world, Indonesia still has scope to improve its productivity, which is currently limited by inconsistency in manual grading through human visual inspection. In this research, an automatic grading machine for oil palm fresh fruits bunch (FFB) is developed based on machine-vision principles of non-destructive analytical grading, using Indonesian Oil Palm Research Institute (IOPRI) standard. It is the first automatic grading machine for FFBs in Indonesia that works on-site. Machine consists of four subsystems namely mechanical, image processing, detection and controlling. The samples used were tenera variety fruit bunches from 7 to 20 year old trees. Statistical analysis was performed to generate stepwise discrimination using Canonical Discriminant with Mahalanobis distance function for classifying groups, and appoint cluster center for each fraction. Results showed adaptive threshold algorithm gave 100% success rate for background removal, and texture analysis showed object of interest lies in intensity within digital number (DN) value from 100 to 200. Group classification of FFBs resulted average success rate of 93.53% with SEC of 0.4835 and SEP of 0.5165, while fraction classification had average success rate of 88.7%. Eight models are proposed to estimate weight of FFBs with average R2 of 81.39%. FFBs orientation on conveyor belt showed no influence on the sorting result, and with examination time of 1 FFB/5 s, machine performs more than 12 tons FFBs grading per hour.