The distribution of tasks between humans and machines in the design of healthcare systems is an important issue for patient safety. This paper presents a usability test performed to compare a semi-automated unit dose system (UDS) with the usual/manual preparation procedure for preparing and administering drugs. The results show that the UDS prevents the frequent administration errors encountered with the usual cabinets and produces a better performance in terms of time for filling the pill dispensers (6.52 sec ± 1.1 vs. 8.5 sec ± 1.5 (t(9, 16) = 3.12, p <.007)). But the results also stress that the UDS takes entire control of the preparation and administration tasks, thus leading to a loss of control of the process by nurses (difficulties experienced by them in resuming their actions, difficulties in memorizing drugs, lack of confidence in the UDS while they "blindly" rely on it). The distribution of tasks between the nurses and the UDS should be modified to give back control to the nurses. Design suggestions were provided in this way. For instance, the UDS may guide the nurses for the drugs localization in the cabinet but leaves the validation of the drugs to the nurses.