One way of assessing the validity of measures of animals‘ fear of people is to examine whether these measures can predict the use of aversive handling techniques by the farmer. This approach has been adopted in a number of studies that have measured dairy cows’ fear of people on dairy farms. Three studies in particular provide data on the correlations across farms between various measures of the fear of dairy cows for people, and measures of the type of handling techniques used. Two studies were done in Australia (Hemsworth et al., 2000—66 farms; Breuer et al., 2000—31 farms) and one in Austria (Waiblinger et al., 2002—30 farms). All of these studies report significant correlations between the measures of the cows’ fear of people and the type of handling technique used, which would seem to support the validity of the measures used. However, the details of the results warn us against too optimistic a conclusion. In Table 2 we present some of the main correlations between handling techniques and cows’ fearfulness, trying as far as possible to present results based on comparable measures. First, the correlations are usually small or moderate. None of the correlations are above 0.5, suggesting that at most only 25% of the variance between farms in handling techniques is related to variance in the measures of cows’ fearfulness. Furthermore, only a proportion of the correlations are significant. The proportion of the correlations presented between the cows’ behaviour and the behaviour of the stockperson which are significant are 4/18 (Breuer et al., 2000), 16/49 (Hemsworth et al., 2000) and 22/70 (Waiblinger et al., 2002). Thus, while we cannot deny that the cows’ behaviour is related to that of the stockperson, it would be difficult to make any precise predictions of what types of handling techniques were used solely on the basis of observing the cows. Second, there are a number of differences between the studies in which exact measures are correlated. For example, Hemsworth et al. (2000) measured flight distance defined as the distance that the cow first moved away from an approaching person. This was done in the paddock (FDP) and in a specially constructed test arena (FDA). Table 2 shows that flight distance in the arena was negatively correlated with the use of positive handling techniques. Thus, a small flight distance would indicate the use of gentle handling. However, this relationship was not found for flight distance in the paddock, which, rather paradoxically, was negatively correlated with the use of aversive handling techniques. Waiblinger et al. (2002) presented a similar measure of flight distance (although they called it approach distance) which was measured in the barn (FDBarn). In contrast to Hemsworth et al. (2000) they reported (Table 2) that this measure of flight distance was both negatively correlated with positive handling techniques and positively correlated with aversive handling techniques. However, this was true only for the handling that was done in the milking parlour: flight distance did not appear to relate to the handling used when collecting cows. All three studies also took measures of the cows’ behaviour during milking by recording each instance of flinching, kicking or stepping during milking. Again, Table 2 shows substantial discrepancies between the studies in how this measure related to the way the animals were handled. It is not our purpose to question the general finding of these studies viz. that use of aversive handling techniques is associated with more fearful cows. This general conclusion seems undeniable. The differences between the studies probably reflect relatively small differences in how measures were taken, how handling techniques were classified etc., and are probably not important in affecting the overall conclusions of the research. However, the fact that there are so many differences between the studies in which specific measures of cows’ fearfulness are predictive of the handling techniques used leads us to question whether we yet have measures of cows’ fearfulness that are sufficiently precise to be used in on-farm welfare audits. For example, it would be difficult to conclude from the three studies whether flight distance or the time in proximity to a person is the best way of measuring cows’ fear of people. The differences between the studies suggest that the exact manner in which handling techniques influence the behavioural response of dairy cows to humans can vary greatly according to the context. This would make it difficult to develop a standardised test that could be used with confidence on all dairy farms. Most importantly for our discussion, the low correlations indicate that any measure of cows’ apparent fearfulness is only a poor predictor of how the cows are handled. Consequently, the validity of the tests in detecting differences between farms in how the animals are handled is low.