Due to their underlying complexity, biomedical systems are typically described by dynamic models with a significant level of uncertainty. Hence, classical control methods cannot be readily applied, as they do not render satisfactory results in terms of performance and even stability. Therefore, this paper proposes a multiple-model adaptive approach to the neuromuscular blockade control problem, based on the recent advances in set-valued observers for model falsification, that explicitly account for unknown parameters in the model. Some of the indistinguishability issues that usually arise in control problems with such levels of model uncertainty are also mentioned, The suggested method outperforms the alternatives in the literature, in terms of reference tracking error and of the number of successful model selections, as illustrated in simulation.