Infrequently omitting a sound from a repetitive sequence elicits the mismatch negativity (MMN) ERP response when the stimulus onset asynchrony (SOA) is less than 200 ms. We contrasted two alternative explanations of omission MMN. (1) Each sound starts a separate temporal integration process. Omissions violate the constancy of the temporal structure within the integration window. (2) Sounds preceding an omission are perceived to be louder than those followed by a sound within the integration period, because omissions allow the full stimulus aftereffect to be included in perceived loudness. We varied the SOA between 117 and 217 ms. For this case, the temporal structure explanation predicts that no MMN will be elicited, whereas the loudness summation explanation predicts that MMN will be elicited. MMN was elicited by tone omissions with random SOA, suggesting that loudness summation plays an important role in the elicitation of omission MMN.