In view of the impossibility theorem of Azriel et al., the Maximum Tolerated Dose (MTD), as currently defined in Phase I and Phase I/II trials, cannot be consistently estimated without making some very strong parametric assumptions. Experimentation carried out at a single dose will fail to provide a consistent estimator of the MTD. We require that information be obtained on a subset of the doses which we describe as the ‘minimal set’. We provide a definition of this set and study its role in semiparametric inference concerning the MTD. Focusing inference on the minimal set, and only indirectly on the MTD, has important consequences. Among these is consistency of estimators for both the minimal set and the MTD, in completely general conditions. Even when the goal is only to estimate the MTD, we do no worse, and often better than other methods and without making any sacrifice in terms of safety for the patient. © 2018, © American Statistical Association and Taylor & Francis 2018.