Feasibility and validity of protocols used to assess welfare were evaluated in two fattening rooms in twenty farms, chosen according to the group size, pen shape, floor and space per animal. The two rooms contained animals at the beginning or at the end of the fattening period, respectively. Behaviour was recorded either through direct observations of the pens (3 times per minute, at 5 min intervals), 15 minutes after entering the room (B1) or after the removal of an object given at the end of the visit (B3). The third protocol consisted of a 2 minute observation inside the pen (B2). To evaluate the human-animal interaction, the time to adapt to the presence of the operator in the corridor (HA1) was measured, as well as the reaction of the group when the operator entered the pen (HA2) or walked slowly through the pen (HA3), the time taken by the first five pigs to approach the operator (HA4) and the reaction of the animals when the operator tried to catch an ear (HA5). Lesion scoring was carried out inside the pen on a sample of 60 individual fatteners or at pen level on most of the pigs from the corridor. The occurrence of the main active behaviours is assessed similarly over time by B1 and B3. Behavioural observations are then possible in an on-farm welfare assessment. Lesion scoring from the corridor tends to under estimate the number of lesions (scratches, tail wounds). The HA3 test appears to be the best one to evaluate the relationship towards humans but has to be validated.