In this paper, we discuss, from an experimental point of view, the use of different control strategies for the trajectory tracking control of an industrial selective compliance assembly robot arm robot, which is one of the most employed manipulators in industrial environments, especially for assembly tasks. Specifically, we consider decentralized controllers such as proportional-integral-derivative-based and sliding-mode ones and model-based controllers such as the classical computed-torque one and a neural-network-based controller. A simple procedure for the estimation of the dynamic model of the manipulator is given. Experimental results provide a detailed framework about the cost/benefit ratio regarding the use of the different controllers, showing that the performance obtained with decentralized controllers may suffice in a large number of industrial applications, but in order to achieve low tracking errors also for high-speed trajectories, it might be convenient to adopt a neural-network-based control scheme, whose implementation is not particularly demanding. Index Terms--Decentralized control, neural networks, selective compliance assembly robot arm robot, trajectory tracking control.