A training teat unit for premature babies has a shield, and a teat connected to the shield and having a mouthpiece. The teat has a base body, which forms a mouth support, wherein the mouthpiece adjoins the base body and narrows in relation to the base body. At least one cavity with at least one suction opening is present in the mouthpiece for the purpose of making available a food droplet or a reward droplet, wherein the cavity is closed except for the at least one suction opening. This training teat unit permits motivation-based and optimal training of the sucking action of a premature baby.
