The simulation of the Atlantic thermohaline circulation (THC) during the Last Glacial Maximum (LGM) provides an important benchmark for models used to predict future climatic changes. This study analyses the THC response to LGM forcings and boundary conditions in nine PMIP simulations, including both GCMs and Earth system Models of Intermediate Complexity. It is examined whether the mechanism put forward in the literature for a glacial THC reduction in one model also plays a dominant role in other models. In five models the THC reduces during the LGM (by 5?40%), whereas four models show an increase (by 10?40%). In all models but one a reduced (enhanced) THC goes with a stronger (weaker) reversed deep overturning cell associated with the formation of Antarctic Bottom Water (AABW). It is found that a major controlling factor for the THC response is the density contrast between AABW and North Atlantic Deep Water (NADW) during the LGM as compared to the modern climate. More saline AABW is consistently found in all simulations, while all models but one show less cooling of AABW as compared to NADW. In five out of nine models a reduced (enhanced) THC during the LGM is associated with more (less) dense AABW at its source region, which in turn is determined by the balance between the opposing effects of salinity and temperature on the density of AABW versus that of NADW. The response in net evaporation over the Atlantic basin is relatively small in most models, so that changes in the freshwater budget are dominated by ocean transports. In only two models is the THC response during the LGM directly related to the response in net evaporation.