We implemented a coarse-graining procedure to construct mesoscopic models of complex molecules. The final aim is to obtain better results on properties depending on slow modes of the molecules. Therefore the number of particles considered in molecular dynamics simulations is reduced while conserving as many properties of the original substance as possible. We address the problem of finding nonbonded interaction parameters which reproduce structural properties from experiment or atomistic simulations. The approach consists of optimizing automatically nonbonded parameters using the simplex algorithm to fit structural properties like the radial distribution function as target functions. Moreover, any mix of structural and thermodynamic properties can be included in the target function. Different spherically symmetric inter-particle potentials are discussed. Besides demonstrating the method for Lennard-Jones liquids, it is applied to several more complex molecular liquids such as diphenyl carbonate, tetrahydrofurane, and monomers of poly(isoprene).