Signal transmission loss and spatial coherence data for source–receiver separations between 100 and 250 km were acquired in the Gulf of Mexico with a calibrated seismic‐streamer measurement system at 400‐m depth, a towed projector at 100‐m depth which emitted 67‐ and 173‐Hz tones, and a moored Webb sound source at 988‐m depth driven at 175 Hz. Range‐dependent bathymetry and sound velocity profiles and other environmental data were measured. The 67‐Hz data showed a persistent sound transmission with a mean of measured range‐averaged loss values (corrected for cylindrical spreading) of 41 dB ranging between 37 and 45 dB; the 173‐Hz data showed several pronounced transmission loss minima with a mean measured range‐averaged loss value of 51 dB ranging between 41 and 60 dB as well as a rapid increase in loss over the slope at ranges greater than 225 km and water depths less than 1.2 km. Slope enhancements were found to be on the order 2–4 dB at 67 Hz and 6 dB at 173 Hz when compared to flat bottom calculations. Pair wise coherence data showed the effect of signal‐to‐noise ratio variations due to multipath interference. Estimates of signal coherence length from the coherent summation of streamer hydrophones yielded coherence lengths ranging between 70 m (8λ) and 300 m (35λ) with an average of 181 m (20λ) at a frequency of 173 Hz (λ=8.67 m). Fast asymptotic coherent and normal mode transmission loss calculations produced results in qualitative agreement with measured data for the deep flat portion of the measurement track when measured geoacoustic profiles or the derived bottom loss curves were used. The results of implicit finite difference parabolic equation calculations were consistent with range‐averaged data for the flat portion of the track as well as on the slope. These results show that if proper descriptions of the subbottom velocity profiles are used, then computations employing either parabolic equation or normal mode techniques provide qualitative agreement with experimental results.