A significant issue in the use of the mismatch negativity evoked potential (MMN) concerns its low signal-to-noise ratio (SNR). One can improve the noise level by increasing the number of samples included in the averaged response. However, improvement achieved in this way assumes that the signal, the MMN, remains stable for extended test times, an assumption which has not been tested. If the MMN is not stable, or exhibits habituation over the test session, then SNR would be adversely affected. MMN response magnitude was measured in 5-min intervals over the course of a test session in response to various speech syllable contrasts. Significant long-term habituation of MMN was observed for all three subject populations tested: young adults, school-age children, and guinea pigs. The time course of the habituation and the stimulus conditions under which it occurs have important implications for research and clinical applications of the MMN. Recording procedures that minimize habituation effects may be used to advantage to improve the signal-to-noise ratio of the MMN.