Mismatch negativity (MMN) is a neurophysiological measure of auditory novelty detection that could serve as a translational biomarker of psychiatric disorders, such as schizophrenia. However, the replicability of its magnetoencephalographic (MEG) counterpart (MMNm) has been insufficiently addressed. In the current study, test–retest reliability of the MMNm response to both duration and omission deviants was evaluated over two MEG sessions in 16 healthy adults. MMNm amplitudes and latencies were obtained at both sensor- and source-level using a cortically-constrained minimum-norm approach. Intraclass correlations (ICC) were derived to assess stability of MEG responses over time. In addition, signal-to-noise ratios (SNR) and within-subject statistics were obtained in order to determine MMNm detectability in individual participants. ICC revealed robust values at both sensor- and source-level for both duration and omission MMNm amplitudes (ICC = 0.81–0.90), in particular in the right hemisphere, while moderate to strong values were obtained for duration MMNm and omission MMNm peak latencies (ICC = 0.74–0.88). Duration MMNm was robustly identified in individual participants with high SNR, whereas omission MMNm responses were only observed in half of the participants. Our data indicate that MMNm to unexpected duration changes and omitted sounds are highly reproducible, providing support for the use of MEG-parameters in basic and clinical research.