Objectives: In the present study, we investigated late-talkers' use of the mutual exclusivity assumption as they inferred the meaning of a novel-word in two different conditions: when a novel word could be the first label of a non-familiar object and when a novel word was the second label of a familiar object. Methods: Fourteen Korean late-talkers age 24 to 35 months and fifteen typically developing peers participated in this study. In Experiment 1, the children were given photos of two objects, one familiar and the other non-familiar, and asked to choose the photo of a novel-word spoken by the examiner. In Experiment 2, the examiner showed the children a photo of a familiar object and used a novel-word (ex, pooty) as its label. Then, the children were shown a photo of that familiar object and a photo of a non-familiar object, and asked to choose which photo corresponded with the novel-word. Results: In Experiment 1, both groups tended to choose non-familiar objects for the novelwords; there was no significant difference between the groups in the frequency of choosing non-familiar objects. In Experiment 2, while the typically developing children tended to map the novel words to the previously shown familiar objects (ignoring the mutual exclusivity assumption), the late-talkers tended to map the novel words to the non-familiar objects. Conclusion: The bias toward mutual exclusivity shown in late-talkers may be useful in learning the first label of an object, but it can restrict acquisition of the various vocabulary related to that object.