Reconstruction of high‐frequency erosion variability beyond the instrumental record requires well‐dated, high‐resolution proxies from sediment archives. We used computed tomography (CT) scans of finely laminated silt layers from a lake‐sediment record in southwest Oregon to quantify the magnitude of natural landscape erosion events over the last 2000 years in order to compare with palaeorecords of climate, forest fire, and seismic triggers. Sedimentation rates were modeled from an age–depth relationship fit through five 14C dates and the 1964 AD 137Cs peak in which deposition time (yr mm‐1) varied inversely with the proportion of silt sediment measured by the CT profile. This model resulted in pseudo‐annual estimates of silt deposition for the last 2000 years. Silt accumulation during the past 80 years was strongly correlated with river‐discharge at annual and decadal scales, revealing that erosion was highly responsive to precipitation during the logging era (1930–present). Before logging the frequency–magnitude relationship displayed a power‐law distribution that is characteristic of complex feedbacks and self‐regulating mechanisms. The 100‐year and 10‐year erosion magnitude estimated in a 99‐year moving window varied by 1.7 and 1.0 orders of magnitude, respectively. Decadal erosion magnitude was only moderately positively correlated with a summer temperature reconstruction over the period 900–1900 AD. Magnitude of the seven largest events was similar to the cumulative silt accumulation anomaly, suggesting these events ‘returned the system’ to the long‐term mean rate. Instead, the occurrence of most erosion events was related to fire (silt layers preceded by high charcoal concentration) and earthquakes (the seven thickest layers often match paleo‐earthquake dates). Our data show how internal (i.e. sediment production) and external processes (natural fires or more stochastic events such as earthquakes) co‐determine erosion regimes at millennial time scales, and the extent to which such processes can be offset by recent large‐scale deforestation by logging. Copyright © 2018 John Wiley & Sons, Ltd. We used computed tomography (CT) scans from a lake‐sediment record in Oregon to reconstruct pseudo‐annual estimates of silt deposition for the last 2000 years, and the relationship with climate, wildfires, and seismic triggers. The frequency–magnitude relationship displayed a power‐law distribution that is characteristic of complex feedbacks and self‐regulating mechanisms. Our data show how internal (sediment production) and external (natural fires or earthquakes) co‐determined erosion over millennia, and the extent to which it is offset by logging in recent times.