The aim of this study was to evaluate the intra-observer repeatability of some animal-related variables which could be used in welfare-assessment protocols at farm level. Recordings were performed on seven dairy farms (four for cattle and three for buffaloes). The animals were observed on three occasions at three-week intervals. The variables collected for each animal were behaviour during milking (stepping and kicking), avoidance distance, lameness, and cleanliness. For each farm and each variable, intra-observer repeatability was computed using the Kendall coefficient of concordance (W). A ratio between the variance of the animal and the sum of the latter with the error variance was also calculated using a model of analysis of variance with one factor (animal), to give a further measure of repeatability. These two methods yielded similar results. In particular, in dairy cattle, a high repeatability for avoidance distance, stepping, lameness and cleanliness was observed, whereas for buffaloes avoidance distance and stepping were the repeatable variables.