Retrospective analyses of fire danger typically use reanalysed data, but its relation to observed fire danger is not well researched. Here we use daily weather observations to calculate fire danger for nine weather stations in Sweden, spanning 1100 km N–S, for the period 1951–2020, making it among the longest series of observed fire danger to date. All sites except the northernmost one exhibited increasing seasonal FWI-metrics over the period and linear trends were statistically significant for three sites. On average, annual peak 7-day moving-average FWI increased by 18% over 70 years. Increasing trends were mostly driven by higher noon-temperature and not by altered precipitation patterns. Further, observed fire danger differed substantially from that based on ERA5 reanalysis data. For FWI > 5, reanalysis FWI-values were on average 25% lower than corresponding observational values. The strength of reanalysis data is to form gridded data using single assimilation schemes against homogeneous model fits and it is not designed to fully represent actual point scale weather. While reanalysis data enables comprehensive geographical analyses, this study shows how it also underestimates peak fire weather in northern Europe. We recommend checking extreme-value bias against point observations in future studies.