We are mounting a hydrophone on autonomous floats capable of drifting below the sound channel and surfacing to communicate data by a satellite link. Using new, intelligent algorithms for the automatic identification and discrimination of seismic phases, we recognize teleseismic arrivals in the presence of local P, S, and T phases, ship and whale noise, and other contaminating factors such as airgun surveys. Our approach combines time-domain methods with spectrogram analysis and with wavelet methods. To maximize battery life, we optimize the efficiency of our algorithms and their numerical implementation. Our algorithms were tested on data from tethered hydrophones from two arrays anchored to the Mid-Atlantic Ridge and the East Pacific Rise. Our prototype device was succesfully tested in a dive to 700 m off the coast of La Jolla. We acquired a valuable 31-h data stream suitable to characterize the ambient noise and were able to identify a number of engineering problems with the hydrophone sensitivity. We expect detecting teleseisms with magnitudes superior to 6.3, and thus adding at least 200 high-quality recordings to the global catalog over the life span of a single instrument—well worth the $15<th>000 manufacturing price and negligible deployment costs on ships of opportunity.