We assessed spatiotemporal patterns in fire severity in 187 fires in the Sierra Nevada, California, USA, between 1984 and 2004. Our work formed part of the US-Forest Service Sierra Nevada Forest Plan Amendment (SNFPA) monitoring program, and sought to document contemporary patterns in severity, to compare these patterns with pre-1985 data which fed into the original SNFPA, to assess to what extent patterns changed over the 20-year period, and to compare current trends with the long-term record. We measured vegetation burn severity by combining remote sensing methods (RdNBR, derived from pre- and one-year post-fire TM imagery) with field data collection (CBI, collected across multiple fires), and stratified our analysis in a variety of ways, including by elevation and dominant vegetation type. Low- and middle-elevation vegetation types characterized historically by high frequency, low severity fire regimes showed strong departures from this presettlement pattern, with significant contemporary skewing toward moderate and high severity fire. Higher elevation vegetation types showed little departure from expected presettlement fire regimes. Jeffrey pine, mixed conifer and white fir vegetation types all experienced increasing trends in the proportion of high severity fire between 1984 and 2004, with the strongest increase occurring in mixed conifer (R2 = 0.829, P < 0.0001, 10-year running average). Mean and maximum size of high severity patches in conifer types also increased significantly over the 21-year period (R2 = 0.544 to 0.5879, P < 0.001), and maximum high severity patch size was positively correlated with fire size (R2 = 0.432, P < 0.001). For most of the 20th century, annual mean and maximum fire size in the Sierra Nevada oscillated in a pseudo-cyclical pattern that is best explained by interdecadal climatic variability. Our data show a strong increase in fire size over the last two decades which appears to be atypical of these trends.