This study presents an optimised set-up for molecular dynamics (MD) simulations of G-protein coupled receptors (GPCR). Such simulations are complicated because (1) the experimental template structure for GPCRs (bovine rhodopsin) is of low resolution, (2) the receptor surroundings are irregular (water exposed loops vs. lipid exposed transmembrane regions) and (3) the protonation and solvation states of the inner core receptor residues are unknown. We compared various simulations of the experimentally derived and refined electron density structure of the seven helical transmembrane protein bacteriorhodopsin (bR) under different MD conditions using AMBER 4.1. Our results demonstrate that the optimal MD set-up with minimal computational effort is a periodic boundary (PB) box containing two water shells solvating the extra- and intracellular loops separated by a vacuum layer surrounding the helical transmembrane (TM) regions. It was found that the vacuum layer and water layers are stable under periodic boundary conditions during at least 1 ns of MD simulation. In this set-up the bR structure is stable without any restraints. The average bR structure during the last 500 ps of the MD run has an excellent RMSD value relative to the original bR structure (RMSD = 1.66 A for the C alpha atoms within the TM domains) and shows a very high helical stability within the TM regions (88.8% helix). The use of this MD set-up for simulations of GPCRs is discussed.