The generation of appropriate behavior is a key competence of automatic vehicles, which significantly determines the safety and efficiency of the overall system. Methods known from classical robotics also formed the basis for the decision-making of automatic vehicles for a long time. Therefore, this chapter first presents rule-based methods. Afterward, graph search methods are described, in which the motion planning is transformed into a graph search. Probabilistic modeling as (PO)MDPs results in methods that make final decisions based on the continuously made observations during execution. Lastly, learning techniques are presented, which have recently been explored and used for behavior generation as well. The architecture of the behavior generation has a significant influence on the achievable scalability, traceability, and safety. As an application example, an arbitrator concept is presented that was implemented in real test vehicles and proved its performance in public road traffic.