Cross-situational word learning, the ability to decipher word-referent links over multiple ambiguous learning events, has been documented across development and proposed to be key to vocabulary acquisition. However, this work has largely focused on learning from one-to-one structure, where each referent is consistently linked with a single label. In contrast, learners can encounter lexical overlap, such as when learning synonyms, which requires learning from structure that presents multiple labels linked with the same thing (many-to-one structure). Are children capable of cross-situational word learning from encounters presenting consistent lexical overlap? How does this ability change with age? Here, we provide a first test of cross-situational word learning from input presenting lexical overlap across age by testing 4-to 7-year-old children (N = 190) and adults (N = 80) on their ability to learn from one-to-one or two-to-one structure. Results showed that adults (Experiment 1) were successful at learning from both types of structure, with an advantage for one-to-one structure. Children (Experiment 2) performed less well than adults overall, and failed to learn from two-to-one structure. With extended training (Experiment 3), older children were successful at learning from two-to-one structure while younger children were not. These results provide the first evidence that cross-situational word learning from one-to-one and two-to-one structure improves over age.