We survey 223 916 auroral records from the Northern hemisphere observed between 1650 January and 2024 July, making full allowance for the secular change in the geomagnetic field. We generate criteria for defining extreme auroral events that are met on 0.015 per cent and 0.023 per cent of nights since 1650 and 1790, respectively. After discussing biases and trends in the data, we compare the event of 2024 May 10–11 with other extreme events and investigate the connections to geomagnetic and sunspot activity. Ranking the events by the lowest geomagnetic latitude from which aurora was observed, the second night of the 2024 May event is shown to be the third most extensive known, the most extensive being 1872 February 4. Allowing for dark adaptation of human vision, we find no evidence that this ranking has been greatly influenced by the increased use of modern digital cameras. We show that the area of the sunspot group from where the causal coronal mass ejection arises (identified by the associated flare) is weakly anticorrelated with the auroral and geomagnetic response; the scatter being large such that, although the 1872 February event arose from a rather small sunspot group, the 2024 May event arose from a large group, as did the 'Carrington Events' of 1859 August/September (ranked 2, 4, and 5). We show that the extreme events all occur during Carrington rotations for which the average open solar flux, $F_S$ exceeds $4{\times }10^{14} ~\mathrm{ Wb}$ but only 3.6 per cent of Carrington rotations when $F_S$ exceeds this value give an extreme event at Earth.