Two experiments investigated the nature of savings for items forgotten from long-term memory (LTM). In Experiment 1, 20 number-noun pairs were learned to a criterion of one errorless trial. The subsequent amount of forgetting from LTM was remarkably low: 10% errors on the retention test after 1 week and 25% after 2 weeks. Items missed on the retention test either remained the same or were changed prior to relearning; same-as-previously missed items were relearned faster than changed previously missed items, implying that ( a) there are savings for individual items forgotten from LTM and ( b) forgetting from LTM is decremental rather than all or none. In Experiment 2, using verbs in sentences, the changed previously missed item was either a synonym, antonym, or verb unrelated to the originally learned verb. There was no relearning difference between these three conditions, suggesting that the partial forgetting residual which benefits relearning is not necessarily semantic in nature.