Introduction: The event-related potential (ERP) reflecting auditory change detection (mismatch negativity, MMN) registers automatic selective processing of a deviant sound with respect to a working memory template resulting from a series of standard sounds. Controversy remains whether MMN can be generated in the frontal as well as the temporal cortex. Our aim was to see if frontal as well as temporal lobe dipoles could explain MMN recorded after pitch-deviants (Pd-MMN) and duration deviants (Dd-MMN). [This lays an important foundation for investigating putative temporal and frontal lobe dysfunction in schizophrenia, where MMN amplitude is reduced in patients with non-paranoid schizophrenia (often predominant negative symptoms) more than in the paranoid forms of schizophrenia (ref 1).] Methods: EEG recordings were taken from 32 sites in 14 healthy subjects during a passive 3-tone oddball presented during a simple visual discrimination and an active auditory discrimination condition. Both conditions were repeated after one month. Hypothesis-led dipole locations were calculated and modeled by BESA and located in Talairach space (Garneron algorhythm). Results: First: The Pd-MMN was larger, peaked earlier and correlated better between test and re-test sessions than the Dd-MMN. Second: Two dipoles in the auditory cortex (superior temporal gyrus) and two in the frontal lobe (left cingulate and right inferior frontal cortex) are described. [Residual variance for the group = 0.87%, and for individuals 0.7-10%]. Third: These dipole sources were similarly placed for Pd-MMN and Dd-MMN, and were well replicated on retest. Some of these data were reported in a preliminary form that contrasted dipole sources with the Negative difference (Nd) waveform found during controlled attentional processes (ref 2). Conclusions: This study confirms interactions between activity generated in the frontal and auditory temporal cortices in automatic attention-like processes that resemble initial brain imaging reports of unconscious visual change detection (ref 3). The lack of interference between sessions shows that the situation is likely to be sensitive to treatment or illness effects on fronto-temporal interactions involving repeated measures.