We present a new approach to cluster simulation as a step toward a molecular theory of vapor phase nucleation. This approach does not involve a cluster criterion which is commonly introduced to determine whether a given molecule in the system belongs to the vapor or to a cluster for any instantaneous configuration of molecules. Instead, the stochastic evolution of the metastable vapor phase was explored in a grand canonical Monte Carlo (MC) simulation aided by the umbrella sampling technique. The physical clusters, i.e., density fluctuations, that lead to nucleation emerge naturally as we attain a coarse-grained description of this stochastic process by introducing proper order parameters, in particular the molecular content and the potential energy of the system suitably chosen in the vapor. Our method also allows an efficient evaluation of the free energy of cluster formation as a function of these order parameters. The method was applied to a Lennard-Jones fluid at temperatures above the bulk melting point. When the physical clusters thus identified were further subjected to a molecular dynamics (MD) simulation, the system was found to evolve along a quasi-continuous trajectory while closely following the valley passing through the saddle point of the free energy surface, thereby demonstrating the relevance of these clusters in the dynamics of nucleation.