Direct fuel injection for Sparc Ignition engines (GDI) has found its entrance in passenger car series production. These engines run lean at low loads and stoichiometric for the other operating area as displayed in figure 1 . In this paper we present a neural net based adaptive injection control strategy for this type of engines. A specific kind of neural network (normalized RBF-net) is used to generate the injection commands for the injection valves. The net is used as a mapping from the driver's wish and operating condition of the engine on injection commands to also achieve a defined (fig. 1) air-fuel-ratio in transient operation. This neural network is trainable online to represent an intelligent feedforward control-structure implementing the artificial intelligence's ability of learning.