We developed a knowledge-based planning method to predict achievable OAR, such as rectum and bladder in cervical cancer, optimization constraints based on patient 3D anatomy similarity. It is a two-step of aligning and matching approach that paired on PTV location and similarity measure on OARs. Using leave-one-out method, we apply this method on a database of 81 cervical IMRT patients. The predicted dose-volume histograms (DVH) parameters were used in treatment planning system to generate a new plan. An example of the new plan was compared with clinical plan by dosimetry goals according to RTOG guidelines.