Abstract The class of hybrid systems describes most technical systems in great detail. However, the respective models and their behavior tend to be very complex. Recently, a new subclass of hybrid automata has been introduced, the Flat Hybrid Automata (FHA) that relies on the concepts of differential flatness for the continuous parts, and strongly connected automaton graphs for the discrete event part, in order to deal with the complexity from a design perspective. Therefore, we introduce in the present paper an approach to reduce the automaton graph of an FHA in a systematic way by removing edges from the adjacency matrix. The main contribution of the paper is twofold: Firstly, based on practical considerations we develop a heuristic algorithm to reduce the automaton graph. Secondly, we present possible ways to include knowledge about the system in the reduction.