Evidence for the assumptions of the salt-advection feedback in box models is sought by studying the Atlantic meridional overturning circulation (AMOC) internal variability in the long preindustrial control runs of two Earth system models. The first assumption is that AMOC strength is proportional to the meridional density difference between the North Atlantic and the Southern Oceans. The model simulations support this assumption, with the caveat that nearly all the long time-scale variability occurs in the North Atlantic density. The second assumption is that the freshwater transport variability by the overturning at the Atlantic southern boundary is controlled by the strength of AMOC. Only one of the models shows some evidence that AMOC variability at 45°N leads variability in the overturning freshwater transport at the southern boundary by about 30 years, but the other model shows no such coherence. In contrast, in both models, this freshwater transport variability is dominated by local salinity variations. The third assumption is that changes in the overturning freshwater transport at the Atlantic southern boundary perturb the north-south density difference, and thus feedback on AMOC strength in the north. No evidence for this assumption is found in either model at any time scale, although this does not rule out that the salt-advection feedback may be excited by a strong enough freshwater perturbation.