Computerized cognitive training is gaining empirical support for use in the treatment of schizophrenia (SZ). Although cognitive training is efficacious for SZ at a group level when delivered in sufficiently intensive doses (eg, 30-50 h), there is variability in individual patient response. The identification of biomarkers sensitive to the neural systems engaged by cognitive training interventions early in the course of treatment could facilitate personalized assignment to treatment. This proof-of-concept study was conducted to determine whether mismatch negativity (MMN), an event-related potential index of auditory sensory discrimination associated with cognitive and psychosocial functioning, would predict gains in auditory perceptual learning and exhibit malleability after initial exposure to the early stages of auditory cognitive training in SZ. MMN was assessed in N=28 SZ patients immediately before and after completing 1 h of a speeded time-order judgment task of two successive frequency-modulated sweeps (Posit Science 'Sound Sweeps' exercise). All SZ patients exhibited the expected improvements in auditory perceptual learning over the 1 h training period (p<0.001), consistent with previous results. Larger MMN amplitudes recorded both before and after the training exercises were associated with greater gains in auditory perceptual learning (r=-0.5 and r=-0.67, respectively, p's<0.01). Significant pretraining vs posttraining MMN amplitude reduction was also observed (p<0.02). MMN is a sensitive index of the neural systems engaged in a single session of auditory cognitive training in SZ. These findings encourage future trials of MMN as a biomarker for individual assignment, prediction, and/or monitoring of patient response to procognitive interventions, including auditory cognitive training in SZ.