The duration-amplitude distribution of volcanic tremor was examined in eight volcanoes and one geothermal area. An exponential model, implying a scale-bound source process, is found to be a better fit to the data than a power law (scale invariant) model. The exponential model well describes tremor associated with magmatic and phreatic eruptions, shallow and deep source regions, and geothermal sources. We tested the exponential model described by: $\ d ( D_{R})= d_{{rm t}}{rm e}^{-lambda D_{R}}$, where d is the duration of tremor greater than or equal to a particular amplitude DR, dt is the total duration of tremor, and the inverse of λ is the characteristic amplitude of the distribution. λ-1 takes on values between 0.003 and 7.7 cm2. Our results show that the characteristic amplitude for eruptive tremor is greater than for noneruptive tremor, that for deep tremor is greater than for shallow tremor, and that for tremor associated with magmatic eruptions is greater than for tremor associated with phreatic eruptions. The exponential scaling of tremor suggests that tremor is not simply composed of a series of low-frequency events closely spaced in time. Further, the exponential scaling requires the source to be scale bound; the amplitude variations of tremor are distributed about a constant characteristic amplitude. We propose that exponential scaling of tremor amplitude is caused by fixed source geometry driven by variable excess pressures. The exponential scaling of tremor demonstrates that tremor source processes are fundamentally different from those for earthquakes.