A computational methodology is presented that is designed to model, at a coarse-grained level, the mesoscale dynamics of fluids and potentially other forms of soft matter. Within a molecular dynamics simulation, "ghost" particles of a specific size, corresponding to the fundamental length-scale of coarse-graining, are used as micro-probes designed to respond to local mesoscale fluid flows and stress gradients. A subsequent coarse-grained model is then developed that incorporates both the coarse-grained mesoscale dynamics and isothermal compressibility of the original microscopic system. The method is applied to water and methanol. A contrast with dissipative particle dynamics (DPD) is also presented. (C) 2004 American Institute of Physics.