Interaction of transmembrane (TM) proteins is important in many biological processes. Large-scale computational studies using coarse-grained (CG) simulations are becoming popular. However, most CG model parameters have not fully been calibrated with respect to lateral interactions of TM peptide segments. Here, we compare the potential of mean forces (PMFs) of dimerization of TM helices obtained using a MARTINI CG model and an atomistic (AT) Berger lipids-OPLS/AA model (AT(OPLS)). For helical, tryptophan-flanked, leucine-rich peptides (WL15 and WALP15) embedded in a parallel configuration in an octane slab, the AT(OPLS) PMF profiles showed a shallow minimum (with a depth of approximately 3 kJ/mol; i.e., a weak tendency to dimerize). A similar analysis using the CHARMM36 all-atom model (AT(CHARMM)) showed comparable results. In contrast, the CG analysis generally showed steep PMF curves with depths of approximately 16-22 kJ/mol, suggesting a stronger tendency to dimerize compared to the AT model. This CG > AT discrepancy in the propensity for dimerization was also seen for dilauroylphosphatidylcholine (DLPC)-embedded peptides. For a WL15 (and WALP15)/DLPC bilayer system, AT(OPLS) PMF showed a repulsive mean force for a wide range of interhelical distances, in contrast to the attractive forces observed in the octane system. The change from the octane slab to the DLPC bilayer also mitigated the dimerization propensity in the CG system. The dimerization energies of CG (AALALAA)3 peptides in DLPC and dioleoylphosphatidylcholine bilayers were in good agreement with previous experimental data. The lipid headgroup, but not the length of the lipid tails, was a key causative factor contributing to the differences between octane and DLPC. Furthermore, the CG model, but not the AT model, showed high sensitivity to changes in amino acid residues located near the lipid-water interface and hydrophobic mismatch between the peptides and membrane. These findings may help interpret CG and AT simulation results on membrane proteins.