The objectives were to evaluate quantitative animal-based measures of sow welfare (lameness, oral stereotypies and reactivity to humans) under commercial farm conditions, and to estimate the influence of housing, sow parity and stage of gestation on the outcome of these measures. Across 10 farms, 311 sows were used. Farms differed in terms of housing design (pen v. stall), space allowance, floor type in stalls (partially v. fully slatted), and feeding system in pens (floor v. trough). Lameness was assessed in terms of gait score, walking speed, stride length, stepping behaviour, response to a stand-up test and latency to lie down after feeding. The presence of oral stereotypies and saliva foam were recorded. Reactivity to humans was assessed by approach (attempt to touch the sow between the ears) and handling tests (exit of the stall for stall-housed sows, or isolation of the animal for pen-housed sows). Only stride length and walking speed were associated with lameness in stall-housed sows (P<0.05 and P<0.01). In stalls, the probability that a sow was lame when it presented a short stride length (<83 cm) or a low speed (<1 m/s) was high (69% and 72%, respectively), suggesting that these variables were good indicators of lameness, but were not sufficient to detect every lame sow in a herd (sensitivity of 0.39 and 0.71, respectively). The stage of gestation and parity also influenced measures of stride length and walking speed (P<0.05). Saliva foam around the mouth was associated with the presence of sham chewing and fixture biting (P<0.05). The probability that a sow presents sham chewing behaviour when saliva foam around her mouth was observed was moderate (63%) but was not sufficient to detect all sows with stereotypies (41%). A high discrimination index was obtained for behavioural measures (aggressions, escapes) and vocalisations during the approach test (stalls: 78.0 and 64.0; pens: 71.9 and 75.0, respectively), the number of interventions needed to make the sow exit the stall during the handling test for stall-housed sows (74.9), and attempts to escape during the handling test for pen-housed sows (96.9). These results suggest that these measures have a good power to discriminate between sows with low and high reactivity to humans. Finally, the outcome of several measures of lameness, stereotypies and reactivity to humans were influenced by the housing characteristics, sow parity and stage of gestation. Therefore, these factors should be considered to avoid misinterpretations of these measures in terms of welfare.