While studying the equational theory of Kleene Allegories (KAl), we recently proposed two ways of defining sets of graphs [BP15]: from KAl expressions, that is, regular expressions with intersection and converse; and from a new automata model, Petri automata, based on safe Petri nets. To be able to compare the sets of graphs generated by KAl expressions, we explained how to construct Petri automata out of arbitrary KAl expressions. In the present paper, we describe a reverse transformation: recovering an expression from an automaton. This has several consequences. First, it generalises Kleene theorem: the graph languages specified by Petri automata are precisely the languages denoted by KAl expressions. Second, this entails that decidability of the equa-tional theory of Kleene Allegories is equivalent to that of language equivalence for Petri automata. Third, this transformation may be used to reason syntactically about the occurrence nets of a safe Petri net, provided they are parallel series.