Abstract: The human-animal relationship is an important factor when considering animal welfare at herd level. In the present study, two behavioural tests for the on-farm assessment of the human-animal relationship at herd level of dairy cows housed in loose housing cubicle systems were evaluated with respect to inter-observer reliability, test-retest reliability, effect of familiarity of test person as well as inter-correlation of the two tests. In a voluntary animal approach (VAA) test, the number of cows and the latencies to approach and touch a stationary test person was measured. In an avoidance (AV) test, the cows' avoidance reactions to an approaching test person were categorised. A first study was carried out in 12 commercial Danish and Austrian dairy farms. On each farm, both behaviour tests were carried out on the same day and repeated within 4-5 days. For each test, cows were tested by both an unfamiliar and a familiar test person (the stock-person) and two observers simultaneously registered the animals' test responses. The inter-observer reliability of both behavioural tests was found to be high (VAA: 2.5-m approach r=0.98 (P<0.001) and touch r=0.97 (P<0.001); AV: Kappa coefficientweighted=0.886 (prevalence index for flight distance>=2 m is 0.636)). The cows at herd level showed shortest latency for touching an unfamiliar test person on the first test day (P=0.006). Further, the AV test had a high test-retest reliability (Kappa coefficientweighted=0.503 (prevalence index for flight distance>=2 m is -0.660)) and results indicated no effect of familiarity of test person (Kappa coefficientweighted=0.463 (prevalence index for flight distance>=2 m is -0.677)). In a second study, the correlation between the two behavioural tests (similar measures) was evaluated. On each of 10 commercial Danish dairy farms with loose housing cubicle systems at four repeated sessions, both behaviour tests were carried out on the same day. For each test cows were tested by the stock-person. The VAA and AV tests at herd level were highly correlated (rs=-0.84; P=0.002). The results suggest that the AV test is valid and applicable for on-farm assessment of the human-animal relationship at herd level. This accounts only partly for the VAA test, which seem to be more unclear regarding motivation for the animals' approach behaviour.