Hubble Space Telescope images of a sample of 285 galaxies with measured redshifts from the Canada–France Redshift Survey (CFRS) and Autofib–Low Dispersion Spectrograph Survey (LDSS) redshift surveys are analysed to derive the evolution of the merger fraction out to redshifts z∼1. We have performed visual and machine-based merger identifications, as well as counts of bright pairs of galaxies with magnitude differences δm≤1.5 mag. We find that the pair fraction increases with redshift, with up to ∼20 per cent of the galaxies being in physical pairs at z∼0.75–1. We derive a merger fraction varying with redshift as ∝(1+z)3.2±0.6, after correction for line-of-sight contamination, in excellent agreement with the merger fraction derived from the visual classification of mergers for which m=3.4±0.6. After correcting for seeing effects on the ground-based selection of survey galaxies, we conclude that the pair fraction evolves as ∝(1+z)2.7±0.6. This implies that an average L* galaxy will have undergone 0.8–1.8 merger events from z=1 to z=0, with 0.5 to 1.2 merger events occuring in a 2-Gyr time-span at around z∼0.9. This result is consistent with predictions from semi-analytical models of galaxy formation. From the simple coaddition of the observed luminosities of the galaxies in pairs, physical mergers are computed to lead to a brightening of 0.5 mag for each pair on average, and a boost in star formation rate of a factor of 2, as derived from the average [O ii] equivalent widths. Mergers of galaxies are therefore contributing significantly to the evolution of both the luminosity function and luminosity density of the Universe out to z∼1.