In the conventional learning control design, a desired output trajectory is specified and an iterative algorithm is implemented to improve the tracking performance as the action is repeated. In this paper, a Model Reference Learning Control scheme (MRLC) is proposed for a class of nonlinear systems that allows the performance of the learning system to be specified by a reference model. An iterative algorithm is designed so that the learning system will eventually follow the desired response specified by the reference model as the action is repeated. This allows extra freedom in learning controller design as the output performance can be specified at the system model leavel in addition to the trajectory level. Design methods for analysing the MRLC system are developed and sufficient conditions for guaranteeing the convergence are derived. Experimental studies on an industrial manipulator are performed to verify the theoretical analysis.