A fully variable valve train significantly increases the degree of freedom of the control of internal combustion engines. Cylinder deactivation, thermal management, alternative combustion strategies, and minimized pumping losses are just a few examples enabled by freely adaptable intake and exhaust valve timings. This paper presents a method to achieve the accurate tracking of load trajectories under stoichiometric conditions. A feedback controller is designed with a mixed-sensitivity H∞ synthesis method. The underlying system plant is modeled by a combination of a mean-value model of the cylinder-internal processes and a neural network to map the correlation between valve timings and cylinder charge. All experiments are conducted on a test bench with a spark-ignited engine equipped with an internally developed fully variable valve train called FlexWork. With this method, a mean absolute error of 0.07bar in indicated mean pressure and of 0.009 in air–fuel equivalence ratio is achieved for the tracking of the reference trajectory. Furthermore, a cost function dependent online optimization of the internal exhaust gas recirculation is conducted without affecting the tracking performance of the load and stoichiometry. Depending on the parametrization of the cost function, nitrogen oxide or hydrocarbon pollutants can be reduced by up to 46% or 17%, respectively.