One hundred and twenty Ss were tested for the effect of natural language mediator (NLM) strength on the retention of NLMs and paired-associate response terms over 1 week. Hypotheses were that stronger NLMs would be retained best, as would the response terms of their pairs. At three different sessions, Ss (1) generated NLMs to 50 CVC pairs, (2) learned 18 of the pairs by a drop-out procedure and had an immediate retention test, and (3) had a 1-week retention test. The main experimental group practiced the pairs along with their NLMs in Session 2, while two groups practiced only the pairs. Two amounts of practice in Session 2 was also a variable. Results were that (1) practice of pairs with their NLMs increased the frequency of NLM recall, (2) NLM and response term recall were both a function of amount of practice, and (3) presence of the NLM during Session 2 practice facilitated recall only in the immediate test.