We suggest a method which enables simulation entities to execute goal oriented behavior planning by dynamic behavior linking. Existing behavior processing methods especially in Semi-Automated Forces (SAF) are mainly based on rather strict plans which are given at the initial stage of simulation, thus they are not effective to cope with contingencies especially at the human in the loop simulation cases as human acts unexpectedly. Even worse, those methods usually suffer from explosion of behavior combination in attempt to describe all possible countermeasures which may be required in various simulation situations, and such combinations may be frequently prone to being inconsistent to the situations or may miss a certain inevitable behavior that an entity must execute. Our method is a way of generating a sequence of behaviors by retrieving behavior linkages from the goal in the manner of back-propagation. In this method we tag behaviors with behavior links (pre/post-conditions), thus simulation entities can be more consistent and more like human since an entity generates behavior links dynamically according to a certain contingency. This paper briefly shows the actual application of the method to the combat model for Korean military and discloses the partial test result.