This paper presents experimental results of robotic excavation based on fuzzy behavior programs. A robotic excavation goal is achieved through several excavation tasks, while a task is completed through sequences of excavation behaviors carried out by an ordered application of primitive, machineexecutable excavation actions. Both tasks and behaviors are specified by finite state machines (FSM) that define all feasible sequences of behaviors or actions. Behavior selection for state transitions in a task FSM is achieved through situation assessment and behavior arbitration. For a behavior FSM, a method of terminating a behavior execution based on the activation levels of its actions is proposed. Since it is almost impossible to infer the exact status of bucket/environment interaction from force/torque data, excavation actions are specified using fuzzy logic rules acquired from human experience and heuristics. Experimental results indicate that the proposed formulation has led to a more efficient execution of excavation tasks than a previous formulation.