The cell membrane exhibits lateral heterogeneity due to the preferential association among the large number of lipid species that constitute the membrane. In particular, the preferential association of cholesterol (CHOL) with saturated lipids into ordered domains has been an area of intense investigation. The large spatiotemporal scales that comprise spontaneous domain formation largely precludes computational investigation via conventional all-atom molecular dynamics. We demonstrate here that molecular coarse-grained (CG) models, obtained from the bottom-up, i.e., via statistical mechanical renormalization of atomistic models, are capable of spontaneous assembly and phase separation for two model raft-like systems, DLiPC/DPPC/CHOL and DOPC/DPPC/CHOL. The resulting bottom-up CG models exhibit spontaneous self-assembly and phase separation and recapitulate the structural correlations of the underlying atomistic models. The accuracy and fast dynamics of these CG models constitute an effective means of bypassing the limited spatiotemporal scales of atomistic simulations. As the first bottom-up CG models of lipid phase separation, the CG models in this work provide an informative analysis for further construction of bottom-up CG models transferable across a range of lipid compositions.