Errors differ in degree of seriousness. We asked whether preschoolers would use the magnitude of an informant's errors to decide if that informant would be a good source of information later. Four- and 5-year-olds observed two informants incorrectly label familiar objects, but one informant's errors were closer to the correct answer than the other's (e.g., one referred to a comb as a brush and the other referred to the same comb as a thunderstorm). When informants had an unambiguous view of the objects, children could identify which informant was closer to being correct, but they did not favor novel labels the "closer" informant later provided. When the informants had an ambiguous view of the objects (e.g., only the handle of the comb was visible), children preferred the novel labels provided later by the "closer" informant. Preschoolers are willing to overlook semantic errors that are close to being correct, but only when there is an understandable reason for the speaker's errors.