We present a new approach to molecular simulation of bubble nucleation. Our approach does not involve any ad hoc criteria to define a bubble for a given instantaneous configuration of molecules. Instead, we explore the stochastic evolution of a system chosen as a small part of the liquid phase by means of an isothermal-isobaric Monte Carlo simulation aided by the umbrella sampling technique. The physical clusters relevant to nucleation, bubbles in the present case, emerge naturally as we attain a coarse-grained description of this stochastic process by introducing proper order parameters, i.e., the volume and the interaction potential of the system. Thus, the concept of cluster commonly employed to describe vapor to liquid nucleation is generalized naturally for the case of bubble nucleation. The method is applied to Lennard-Jones fluids to evaluate the free energy of bubble formation under a moderate negative pressure. The interaction potential plays a similar role to that in vapor to liquid nucleation in that it characterizes the spatial extent of the bubble. There is thus a unity in free energetics of vapor to liquid nucleation and of bubble nucleation.