Abstract: Broadband underwater acoustic recordings from the McMurdo Oceanographic Observatory mooring near the seaward terminus of the McMurdo Station seawater intake jetty. An omnidirectional Ocean Sonics icListen hydrophone (SB2-ETH, SN 1713) recorded continuously at 512 kilosamples/second (256 kHz Nyquist frequency; 24 bit) for 2 years. The hydrophone was mounted vertically on a steel strut (insulated with rubber sheet) at about 70 cm above the mud/gravel seabed at 21m deep, with the sloping 45° rubble face of the jetty just behind the hydrophone. Temporal coverage is >90%, with gaps and truncated files arising due to network and power outages and software bugs. The audio recordings are 10 minute WAV files, compressed using the lossless FLAC code (Free Lossless Audio Codec, xiph.org; about 33MB of data/minute compressed; 100MB/min uncompressed). The hydrophone was under thick (to 3 m) sea ice cover for the majority of the dataset. The majority of the recorded biological sounds were produced by Weddell seals. Orca were present intermittently (~10 days total) in January-March in both summers. Known non-biological sounds include irregular low-intensity, broad-spectrum clicks and cracks from the sea ice cover, occasional wind noise, a 1.5-s gurgle with components to 200kHz every 90s from the CTD’s pump, a broad-spectrum mechanical sound for 3 min every 4 h from the observatory's underwater camera cleaning system, low-intensity whines (about 18, 58, 83, and 130 kHz, though variable over the dataset) thought to be from the station seawater pumps (>100 m away within the jetty’s well casing), and intermittent noises from tracked-vehicles and helicopters (September–February), SCUBA divers (October–December), and ships (January). Given hosting limitations, only every 6th file (roughly 10min/hour) has been archived here. Additional data can be obtained by contacting the primary author of the dataset, who will maintain it for as long as possible. Audio spectrogram images (PNGs) at three frequency ranges (three stacked panels per image, upper limits of 2.5, 25, and 256 kHz) from the entire dataset (all data, not subsampled) are also archived separately.