During the summers 2015, 2016 and 2017, we carried out some gravity campaigns at the Askja Volcano, in collaboration with the University of Iceland and the financial support of FUTUREVOLC, GSRG, COMET, RAS and EUROVOLC. In 2015, we added five new benchmarks to the existing gravity network (MYV1, MYV2, CASK, STAM and VATN), and I redefined the benchmark at DYNG. Each campaign was carried out over 4-5 days, using the Scintrex CG-5 gravimeter No. 968, owned by the University of Iceland. For each survey, I was accompanied by a co-worker, Stephanie Dumont (2015, 2016) and Andy Hooper (2017), to help handling the gravimeter smoothly. Each measurement consisted of a series of 5 1-min samples, automatically recorded successively, and a despiking filter was applied during the recording. Benchmarks were usually reoccupied twice, ideally at different times of the day, but when necessary the two measurements were carried out one after the other, with a break of 10-15 minutes in between. Vibration noise was minimised by sheltering the gravimeter and remaining static during the 5 minutes of measurement. The files provided here list the raw data recorded per day during each survey (YYYYMMDD.csv). In each file, the column “GRAV” provides the gravity measurement per site occupation, with “SD” being the associated standard deviation. It is worth noting that an automatic Earth tide reduction has been applied to the measurements, but it can be manually cancelled using values in the column “TIDE”.