Behavioural assessments used by shelters to determine adoption suitability of dogs should ideally provide an efficient and accurate profile of each dog's behavioural characteristics. However, a scientific review of assessments currently used in Australia revealed a lack of standardisation and validation. To address this issue the B.A.R.K. protocol was developed; incorporating the best features of existing protocols into 12 subtests which aim to assess how a dog might behave in real life situations. Five behavioural attributes (anxious, non-compliant, fearful, non-friendly and active) were scored on a scale from 0 (behaviour absent) to 10 (behaviour extreme) for all subtests. To assess the reliability of the B.A.R.K. protocol, 48 dogs in a Victorian animal shelter were assessed on two consecutive days. The first assessment was scored independently by two observers. Pearson correlation analyses revealed that most measures of inter-rater reliability were statistically significant (p<.01) but ranged from weak to strong (r=.38 to .99). A second assessment was conducted 24 hours later and scored by one of the original observers. Again, Pearson correlation analyses revealed that measures of test-retest reliability were typically significant (p<.01) but ranged from weak to strong (r = 0.32 to 1.00). Measures of Compliance and Friendliness were most reliable (r = 1.00, n = 31, p<.001), with a perfect correlation for the 'resource guarding wet food' subtest. Findings that some correlations were not significant, and that many were weak to moderate in strength, raise concerns about whether some canine behavioural traits can be reliably measured in the shelter environment. This issue is critical having implications for both public safety and dog welfare; behavioural assessments, whether valid or not, are currently used to determine which dogs are suitable as pets, and which should be euthanased rather than rehomed. Further research is underway to investigate the validity of the B.A.R.K. protocol.