This study addresses a problem of learning multiple continuous trajectories by means of recurrent neural net works with (in general) time-varying weights. The learning task is transformed in to an optimal control problem where both the weight sands initial network state to be found are treated as controls. Based on a variational formulation of P on try agin's maximum principle, a new learning algorithm is proposed which generalizes that given in [11]. Under reasonable assumptions, its convergence is also discussed. A numerical example of learning a two-class problem is presented which demonstrates the efficiency of the approach proposed.