Tsunami earthquakes are highly destructive. The runup produced by their associated tsunami is much larger than expected for most megathrust events of the same magnitude. These rare events have modest magnitudes, but rupture the shallowest portion of a subduction zone megathrust with exceptionally large seafloor displacements. Reliable and unsaturated estimates of the moment magnitude of these events are not generally available before their tsunamis inundate. Furthermore, the event magnitude alone is not an indication of the runup that these tsunami earthquakes produce. Thus it remains challenging to discriminate these events as tsunami earthquakes in time to issue local tsunami warning. Previous teleseismic observations found that tsunami earthquakes radiate seismic waves weakly, and should therefore not be strongly felt in the near field. To date no near‐source seismic recordings of these events exist that confirm this. I will present on an analysis of near‐field records of a dataset of subduction zone earthquakes, including the 2010 M7.8 Mentawai, Indonesia tsunami earthquake. The previously unavailable records of the Mentawai event show remarkably weak shaking, in comparison to both the other events as well as to estimates from ground-motion models. The M7.8 event shakes like a M6.3 event, but inundates like a M9 event. This is strong evidence that this earthquake does indeed have a weakly radiating or inefficient source process, in spite of its large slip. These findings imply that because other tsunami earthquakes are inferred to have similar source properties, our methodology should be applicable to other events.