Extinction of learned fear is facilitated by the partial NMDA agonist D-cycloserine (DCS). However, some studies suggest that the involvement of NMDA in learning differs depending on whether learning is for the first or second time. The current study aimed to extend these findings by examining the role of NMDA in extinction for the first and the second time. Specifically, the present series of experiments used Pavlovian fear conditioning and extinction paradigms to compare the effect of DCS on extinction of fear to a light CS the first and second time around. As found previously, DCS facilitated extinction of learned fear (Experiment 1). A novel finding, however, was that DCS did not facilitate the re-extinction of fear to this same CS following retraining (Experiments 2A and 2B). Finally, it was demonstrated that the transition from NMDA-dependent to NMDA-independent extinction was stimulus specific (Experiment 3). That is, rats were first trained to fear a CS (light); this fear was then extinguished. Following this, rats were then retrained to fear the same CS (light) or a new CS (white noise). When given a second extinction session, DCS was found to facilitate extinction of the new CS but not the original CS. The results of this series of experiments suggest that the role of NMDA in extinction depends on whether extinction is new learning (first extinction) or retrieval of a previous extinction memory (re-extinction).