Reaction systems are a model of interactive computation, where the interaction between a system — itself built up of a number of reactions — and its environment is modelled through context sequences provided by the environment. The standard execution semantics of reaction systems is synchronous, i.e., at each computational step all the enabled reactions are executed. In this paper, we ‘de-synchronise’ such an execution model by allowing only a subset of enabled reactions to be executed. We then study the resulting asynchronous model assuming two fundamental execution policies. The first one allows any subset of reactions to be executed, and the second one draws each subset from a pre-defined pool. We also introduce and discuss the notion of persistence of reactions and sets of reactions in the resulting models of asynchronous reaction systems. In particular, we demonstrate that reaction persistence can be implemented.