The dominance structure in an established group of sheep is often difficult to determine in the field, because of the low frequency of agonistic interactions. One solution to this problem is to test each pair separately in a competitive situation. The most frequently used test involves competition for food after a specific period of food deprivation, but to date, there is no agreement as to the optimum length of time the animals should go without food. The present experiment compared competitive behaviour of sheep at different lengths of time after a meal. The experimental animals were a group of 10 1- year old female Scottish Blackface sheep (45 dyads). The animals had access to dry grass daily, from 07:00 to 08:00 h and from 19:00 to 20:00 h. They were tested in pairs for 5 minutes immediately after and 2, 4, 6 and 8 h after the end of the morning meal. In the test arena a feed hopper provided access to feed to only one sheep at a time. Aggressive interactions such as blocking, threats and butts, as well as non-aggressive interactions (nudges), the time spent at the feeder, and the number and type of displacements (aggressive or non-aggressive) were recorded. The number of displacements was also recorded in the home pens during ten 45-minute periods of group feedings. Blocking behaviour showed the clearest difference between the two sheep in a pair and was therefore, used as the main indicator of dominance. Its occurrence increased as the time since the morning meal was increased and decreased in the course of the 5 minute tests. Consistency between pair-tests and group observations was greatest in the test 8 h after the previous meal, but the difference in consistency between the 2 and 8 h tests was small. This experiment suggest that a period of at least 2 h of food withdrawal is necessary to generate sufficient motivation for a competition test and to mask differences in feeding motivation between individuals. While the performance of dominance-relevant behaviour increased with time since the last meal, a period of 4 h may be sufficient to provide high-quality data. We also suggest that the test duration can be reduced to 2 minutes.