A tsunami coda lasting more than 5 days was clearly recorded across the Pacific Ocean following the 2011 Tohoku‐Oki earthquake. We found that the coda energy per unit area (elapsed time of 60–96 hr from the earthquake origin time) does not show a systematic variation with respect to the travel distance at the stations located at sea depths deeper than 2000 m, whereas the maximum amplitude near the source is significantly larger than those of the far‐field stations. This feature is similar to that of a seismic coda in the sense that the coda energy is distributed uniformly in space. The average tsunami coda decays with a decay time of ∼24 hr. When we carefully examined the tsunami waveform near Japan, a large‐amplitude wave was identified at an elapsed time of ∼48 hr, which corresponds to the round‐trip travel time for waves reflected from Chile. To investigate the excitation mechanisms of the tsunami coda, we conducted a tsunami simulation using a global model. Although linear long‐wave tsunami equations were found to overestimate the observed coda amplitude, the equations including an intrinsic energy loss could successfully simulate the observed coda decay if a bottom friction coefficient of 10−5 was assumed. This indicates that not only wave scattering but also intrinsic energy loss plays an important role in the tsunami coda excitation and decay process.