Dynamic protocol recovery tries to recover a com- ponent's sequencing constraints by means of dynamic analysis. This problem has been tackled by several automaton learning approaches in the past. These approaches are based on the sequence of component method invocations only. We introduce a new dynamic protocol recovery tech- nique based on object process graphs. These graphs con- tain information about loops and the context in which methods are being called. We describe the transforma- tion of a set of these graphs to a protocol automaton. The additional input, compared to the sole sequence of method calls, results in a more detailed protocol. In a case study, we compare the resulting protocol automata of our approach to those of several existing automaton learning approaches.