Most gait-scoring scales for pigs have a limited number of categories, supposedly to improve repeatability. However, reducing the number of categories could lead to loss of information if the observers' discriminative capacities are underused. With a recently estimated within-herd prevalence of sow lameness of 8.8% to 16.9% in the European Union and the associated losses, the availability of reliable tools for the timely detection of initial cases warrants attention. This study investigated the intra- and inter-observer repeatability (intra-OR and inter-OR) of three gait-scoring scales for sows: a continuous 'tagged' visual analogue scale (tVAS, measured in mm), a 5-point and a 2-point ordinal scale (5P and 2P), all with the same descriptors. Veterinary medicine students (n=108) were trained to use the scales and then asked to score 90 videos (30 per scale) of sows with normal and abnormal gait. Thirty-six videos were shown once and 18 were randomly shown three times, of which one mirrored horizontally. The students' opinions on the scales were also collected. Intra- and inter-OR were higher with the tVAS than the 2P scale (inter-OR: 0.73 v. 0.60; P<0.05. Intra-OR: 0.80 v. 0.67; P<0.05). Intra-OR was higher with the 5P (0.81) than the 2P scale (0.67; P<0.05). For all three scales, repeatabilities were lower (P<0.05) for non-lame sows (gait score of ⩽45 mm on the tVAS) than for sows showing some signs of lameness (gait score>45 mm). Video order (first 45 v. last 45 clips), mirroring, users' opinions on the scales, and previous declared experience in handling pigs or scoring lameness in other species had no effect on repeatabilities. Correlations between the students' and experts' scores were high (tVAS=0.92; 5P=0.91; 2P=0.88) but the association for the 2P was not linear and the frequency distribution showed lower correlations for a group of students. This study confirms recent evidence that it is possible to design high-resolution gait-scoring scales that do not reduce observer repeatability. Visual gait-scoring scales with fewer than five categories are likely to entail loss of information on lameness in individual sows.