This paper deals with the study of behavioral equivalences between labeled (place/transition) Petri nets preserving the causal relationship among transition firings. This is achieved by defining a very slight modification in the conventional Petri net formalism, leading to associating an additional information to each token of the net. Following this approach, bisimulation equivalences are defined within a partial order framework, in a way similar to Park's notion of bisimilarity originally defined for interleaving based models. >