The aim of this study was to evaluate the repeatability of some animal related variables, which could be used in proto-
cols developed for assessing animal welfare at farm level. Recordings were performed in seven dairy farms (four for cat-
tle and three for buffaloes). The animals were observed on three occasions at three-week intervals. The variables col-
lected for each animal were the following: behaviour during milking (stepping and kicking), avoidance distance, lame-
ness and cleanliness. For each farm and each variable repeatability was computed using the Kendall coefficient of con-
cordance (W). In buffalo farms avoidance distance may be considered highly reliable (W > 0.64), whereas in dairy cat-
tle its reliability ranged from medium (W = 0.43 to 0.59) to high (W = 0.64). Behavioural recordings at milking showed
that the reliability of stepping was either medium or high for both buffaloes and cattle (W = 0.51 to 0.66 and W = 0.52
to 0.76 for buffaloes and cattle, respectively). Conversely, kicking was less reliable. In cattle farms the reliability for
cleanliness ranged from medium (W = 0.51) to high (W = 0.62 to 0.71), whereas, it was not reliable in the sole buffalo
farm where this variable was monitored. In cattle farms, the concordance for lameness score was high in two farms (W
= 0.62 and 0.66) and moderate in one farm (W = 0.43), whereas no animals displayed lameness in the fourth farm. In
all buffalo farms no animals showed lameness. For each species, the Kruskal-Wallis analysis of variance with one factorshow abstract