In this study, we propose a new global procedure to perform optimization of semiempirical intermolecular potential parameters on the basis of a large reference database. To obtain transferable parameters, we used the original method proposed by Ungerer ͓Ungerer et al., J. Chem. Phys. 112, 5499 ͑2000͔͒, based on the minimization of a dimensionless error criterion. This method allows the simultaneous optimization of several parameters from a large set of reference data. However, the computational cost of such a method limits its application, because it implies the calculation of an important number of partial derivatives, calculated by finite differences between the results of several different simulations. In this work, we propose a new method to evaluate partial derivatives, in order to reduce the computing time and to obtain more consistent derivatives. This method is based on the analysis of statistical fluctuations during a single simulation. To predict equilibrium properties of olefins, we optimize the Lennard-Jones potential parameters of the unsaturated hydrocarbon groups using the anisotropic united atoms description. The resulting parameters are consistent with those previously determined for linear and branched alkanes. Test simulations have been performed at temperatures ranging from 150 to 510 K for several ␣-olefins ͑ethylene, propene, 1-butene, 1-pentene, 1-hexene, 1-octene͒, several ␤-olefins ͑trans-2-butene, cis-2-butene, trans-2-pentene͒, isobutene, and butadiene. Equilibrium properties are well predicted, and critical properties can be evaluated with a good accuracy, despite the fact that most of the results constitute pure predictions. It is concluded that the AUA potential, due to a relevant physical meaning, can be transferred to a large range of olefins with good success.