The individual activity level of pregnant sows can be estimated using accelerometers attached to the ear, but these sensors are fragile and expensive. They can be replaced by manual behavioural analysis of continuous video recordings (“focus”), which is timeconsuming, or analysis of still images selected at regular points in time (“scan”). However, do these two video methods yield the same results when estimating the level of physical activity of a sow? What behaviours (e.g., position, type of activity) should be considered to define this activity? The behaviour of 22 pregnant sows was analysed manually using the focus and scan methods from videos of 2 hours recorded over 2 days (i.e., 30 and 103 days of gestation, respectively). The focus method recorded the duration of each type of activity (eating, drinking, exploring, or having social contact), while the scan method recorded, every 5 minutes, the sow’s posture (standing, sitting, or lying) and type of activity. Three activity levels (high, medium, or low) were assigned to each sow based on its types of activity and postures. The results indicated that the focus and scan methods estimated the same activity levels for the sows (test de Khi2, 80 % vs 56 % similarity, P < 0.01). Activity groups based on types of activity or postures alone also contained the same sows (Khi2, 55-65 %, P < 0.01). Finally, with the scan method, the groups based on the eating and drinking activities alone were similar to those based on all types of activities (Khi2, 63 % vs 72 % similarity, P < 0.01). The activity level can thus be estimated using only data automatically recorded every 5 minutes. To generalize these results, this study needs to be repeated over a full day.