This paper presents the iterative learning control of an electro-hydraulic fully flexible engine valve actuation system. The specific camless system has a unique hydro-mechanical feedback mechanism that simplifies the external control to the choice of triggering timings for three two-state valves. All the critical parameters describing the engine valve event, i.e. lift, duration, timing, and seating velocity, can be continuously varied by controlling these timings. Initial testing of a prototype experimental setup reveals that the performance of the system (transient tracking and steady-state variability) is influenced purely by the state of the system when the internal feedback mechanism is activated. This feature, along with the cyclic nature of the engine valve operation, motivates the development of a iterative-learning-based feedback and feed-forward controller to identify and set the optimal operating point in real time using the output of the previous cycle and the desired performance. The learning control implementation presented here is unique in that, instead of calculating a control signal (sequence) for each cycle, it sets the triggering timings for each of the on-off valves, which directly affect the initial conditions for the internal feedback loop. Experimental results demonstrate that the controller is able to minimize lift and closing time errors while satisfying the seating velocity constraint even during aggressive transient operation.