Horses can be ridden with or without a bit. Comparing the behaviour of the same horse in different modes constituted a ‘natural experiment’. Sixty-nine behaviours in 66 bitted horses were identified as induced by bit-related pain and recognised as forms of stereotypic behaviour. A prototype questionnaire for the ridden horse was based on 6 years of feedback from riders who had switched from a bitted to a bit-free bridle. From a template of 69 behavioural signs of pain derived from answers to the questionnaire, the number of pain signals shown by each horse, first when bitted and then bit-free, was counted and compared. After mostly multiple years of bit usage, the time horses had been bit-free ranged from 1 to 1095 days (median 35). The number of pain signals exhibited by each horse when bitted ranged from 5 to 51 (median 23); when bit-free from 0 to 16 (median 2). The number of pain signals for the total population when bitted was 1575 and bit-free 208; an 87% reduction. Percentage reduction of each of 69 pain signals when bit-free, ranged from 43 to 100 (median 87). The term ‘bit lameness’ was proposed to describe a syndrome of lameness caused by the bit. Bit pain had a negative effect on proprioception, i.e. balance, posture, coordination and movement. Only one horse showed no reduction in pain signals when bit-free. The welfare of 65 of 66 horses was enhanced by removing the bit; reducing negative emotions (pain) and increasing the potential to experience positive emotions (pleasure). Grading welfare on the Five Domains Model, it was judged that – when bitted – the population exhibited ‘marked to severe welfare compromise and no enhancement’ and – when bit-free – ‘low welfare compromise and mid-level enhancement.’ The bit-free data were consistent with the ‘one-welfare’ criteria of minimising risk and preventing avoidable suffering.