The objective of the present study was to analyse the occurrence of undesirable behaviour traits registered on sheep dog trials in order to evaluate the importance of genetic and environmental sources of variation and to draw conclusions for breeding purposes. The data analysed consisted of 2745 test results recorded at 48 sheep dog trials carried out in Germany from 1994 to 1998, which were attended by 337 sheep dogs. Variance components of undesirable behaviour traits were estimated applying Restricted Maximum Likelihood methods. Additive genetic effects, permanent environmental effects of the animal and the effect of the handler were treated as random factors. Additionally, the linear multivariate animal model included the fixed effects of the age of the dogs at the sheep dog trial, sex, the level of difficulty of the exercises as well as the event itself, the starting number of the dog, the number of the dogs' tests at the particular event and the number of dogs presented on sheep dog trials by their handlers. The inbreeding coefficient was regarded as a linear covariate. The analyses were performed using all sheep dog trial classes and for each of the three classes separately. Age of the dog and starting number did not explain a significant proportion of variance for traits analysed, whereas the event of the sheep dog trial and partly the number of dogs per handler were of significant importance. The estimated heritabilities for the undesirable behaviour traits ranged from h2 < or = 0.001 to h2 = 0.07 with standard errors in the range between 0.001 and 0.06. The possibilities to select against undesired behaviour traits appear to be rather limited given the heritability estimates and the low number of progeny.