It has been argued that diapycnal mixing has a strongly stabilizing role in the global thermohaline circulation (THC). Negative feedback between THC transport and low-latitude buoyancy distribution is present in theory based on thermocline scaling, but is absent from Stommel’s classical model. Here, it is demonstrated that these two models can be viewed as opposite limits of a single theory. Stommel’s model represents unlimited diapycnal mixing, whereas the thermocline scaling represents weak mixing. The latter limit is more applicable to the modern ocean, and previous studies suggest that it is associated with a more stable THC. A new box model, which can operate near either limit, is developed to enable explicit analysis of the transient behaviour. The model is perturbed from equilibrium with an increase in surface freshwater forcing, and initially behaves as if the only feedbacks are those present in Stommel’s model. The response is buffered by any upper ocean horizontal mixing, then by propagation of salinity anomalies, each of which are stabilizing mechanisms. However, negative feedback associated with limited diapycnal mixing only prevents thermohaline catastrophe in a modest parameter domain. This is because the time-scale associated with vertical advective-diffusive balance is much longer than the time required for the THC to change mode. The model is then tuned to allow equilibrium THC transport to be independent of the rate of mixing. The equilibrium surface salinity difference controls the classical THC-transport/salinity positive feedback, whereas the equilibrium interior density difference controls the mean-flow negative feedback. When mixing is strong, unrealistic vertical homogenization occurs, causing a convergence in surface and interior meridional gradients. This reduces positive feedback, and increases stability, in the tuned model. Therefore, Stommel’s model appears to overestimate, rather than underestimate, THC stability to high-frequency changes in forcing.