AbstractWe estimate and model the normalized moment rate power spectrum of large slow slip events in Cascadia. We estimate the spectrum using data from GPS‐derived slip inversions, borehole strain records, and beamforming‐based tremor amplitudes. The normalized power spectrum initially decreases with frequency but then may flatten at periods of 1 to 10 days before decaying as frequency at higher frequencies, where nm is between 1.1 and 1.4 when estimated from tremor and between 0.4 and 1.5 when estimated from strain. We explore one way to understand the observed spectrum: by modeling a month‐long slow slip event as the sum of a steady background moment rate and a population of subevents. The subevents represent the wide variety of observed slow earthquakes, ranging from 0.5‐s‐long tremor to 3‐hr‐long rapid tremor reversals. We parameterize the subevents' magnitude distribution and moment‐duration scaling, and we examine how the subevent population determines the slow slip spectrum. There are not enough data to rigorously test the subevent model, but we show that the data are consistent with a single continuum of slow earthquakes whose moments scale linearly with their duration, as has been proposed previously.