Individuals in anxious states pay more attention to threats than calm individuals, and this response is termed attention bias. In the current study a threat perception test was validated using pharmacological models to induce and reduce anxiety in Merino sheep. Sixty sheep received one of the following treatments (n=20/treatment): 1. control (saline), 2. diazepam (0.1 mg/kg, i.v., anxiolytic), or 3. m-CPP: 1-(m-chlorophenyl)piperazine (2 mg/kg, i.m., anxiogenic). Thirty minutes after treatment, each sheep entered an arena (4 x 4.2m) for the 3 minute test. The arena contained hay in the centre and a dog (sitting quietly) was visible to the sheep for 3 seconds before the window was closed. Latency to eat, time spent eating, vigilance behaviour (defined as having the head at shoulder height or higher) and zones crossed in the arena were measured in the test period. There was a treatment difference in vigilance behaviours (P<0.01).  Sheep receiving m-CPP spent more time vigilant than controls (95 ± 3.9% and 77 ± 3.9% respectively). Diazepam treated sheep spent less time vigilant (64 ± 3.9%) compared with the other treatments (P<0.01). The m-CPP group had a higher latency to eat (180 ± 11.7s; P<0.01) than control or diazepam groups (49 ± 11.7s and 44 ± 11.7s, respectively) and spent less time eating (P<0.01). The diazepam and control groups did not differ in latency to eat however the diazepam group spent more time eating overall (P<0.01). There were no differences in zones crossed between the groups (P=0.63). The differences in vigilance and latency to eat between the treatment groups provide evidence that the threat perception test paradigm is measuring states of anxiety in sheep. This novel attention bias test may be useful to assess negative affective states in sheep which could be used in welfare assessment.