We investigated late and cognitive (mismatch negativity, P300) auditory potentials in 14 children with cochlear implants between the ages of 4 and 12 years. Length of cochlear implant use ranged from 7 to 84 months. Three types of stimulus contrasts were used: (1) a loudness contrast consisting of a 1500 Hz tone burst presented at 75 (standard) and 90 dB sound pressure level (deviant); (2) a frequency contrast consisting of a 1500 Hz tone burst (standard) and a 3000 Hz tone burst (deviant) presented at 80 dB sound pressure level; and (3) a speech contrast consisting of "heed" (standard) and "who'd" (deviant) delivered with a roving loudness paradigm involving a randomized variation of the levels of the standard and deviant stimuli. Latencies and amplitudes of components N1, P2, N2, and P3 and a mismatch negativity were measured. Overall, there were very few missing or unidentifiable components. P3 and mismatch negativity components were identified for all subjects and all stimuli. The latencies of most components were affected by stimulus type. There was a trend for longer latencies for the speech contrast compared with the loudness or frequency contrasts. This may be a reflection of the increased processing time required for the speech stimuli because of its higher complexity. There were several significant correlations between speech recognition and cognitive evoked potential latencies. These results indicate that the clinical use of cognitive evoked potentials in children with cochlear implants is feasible and informative.