Electrocochleography is increasingly being utilized as an intraoperative monitor of cochlear function during cochlear implantation (CI). Intracochlear recordings from the advancing electrode can be obtained through the device by on-board capabilities. However, such recordings may not be ideal as a monitor because the recording electrode moves in relation to the neural and hair cell generators producing the responses. The purposes of this study were to compare two extracochlear recording locations in terms of signal strength and feasibility as intraoperative monitoring sites and to characterize changes in cochlear physiology during CI insertion. In 83 human subjects, responses to 90 dB nHL tone bursts were recorded both at the round window (RW) and then at an extracochlear position-either adjacent to the stapes or on the promontory just superior to the RW. Recording from the fixed, extracochlear position continued during insertion of the CI in 63 cases. Before CI insertion, responses to low-frequency tones at the RW were roughly 6 dB larger than when recording at either extracochlear site, but the two extracochlear sites did not differ from one another. During CI insertion, response losses from the promontory or adjacent to the stapes stayed within 5 dB in ≈61% (38/63) of cases, presumably indicating atraumatic insertions. Among responses which dropped more than 5 dB at any time during CI insertion, 12 subjects showed no response recovery, while in 13, the drop was followed by partial or complete response recovery by the end of CI insertion. In cases with recovery, the drop in response occurred relatively early (<15 mm insertion) compared to those where there was no recovery. Changes in response phase during the insertion occurred in some cases; these may indicate a change in the distributions of generators contributing to the response. Monitoring the electrocochleography during CI insertion from an extracochlear site reveals insertions that are potentially atraumatic, show interaction with cochlear structures followed by response recovery, or show interactions such that response losses persist to the end of recording.