Abstract Six sheep were trained, over a period of 11 weeks, to work for a reward of freshly cut clover by walking along a variable length, U-shaped race. Training included habituation to the presence of an observer and to isolation from conspecifics. Following training, motivation to eat clover was tested by varying the work (distance walked) required to access a 5 g (dry matter) clover reward. Each sheep was tested using three different race lengths (3, 8 or 20 m) in random order on three different days. In each test, the sheep were allowed up to 1 h to access the reward 25 times. The sheep could only obtain clover in the reward area but could graze grass at all other times. There was a linear decline in the number of visits to the reward area within 1 h ( P = 0.05) as the distance walked increased. Distance walked did not significantly affect walking speed in the race ( P = 0.10). The latency between visits showed a linear decline as number of visits increased ( P = 0.01). The total amount of clover consumed and the time spent in race transit and reward area residency were used to calculate effective clover intake rates for each treatment (5.1, 4.4 and 3.7 g DM min −1). Theoretical calculations indicated that there would be a net energy benefit from obtaining the clover reward at the two shorter distance treatments, but that using the system at the longest distance would only yield a very small energy benefit.