The behaviour of animals towards humans is widely used to assess human–animal relationship and welfare in livestock. The aim of this study was to develop an approach test for dogs that is feasible in a surveillance setting, shows stability over a given time period and has good between-experimenter repeatability. To assess validity of the test, relationships between dog behaviour and attitudes of shelter staff to dogs, dog care and dog handling were explored. The test consisted of approaching the front of a dog's kennel in a non-threatening manner. Dogs were categorized as “contact possible” if they approached and explored the experimenter and as “no contact possible” if they ignored, attacked, or barked or growled continuously. We tested 520 dogs in 29 shelters and a mean percentage of 76 ± 19% (mean ± S.D.) of dogs per shelter were categorized as “contact possible”. Stability over time (mean number of days between two visits: 58) was tested on shelter level in nine shelters and resulted in a rs = 0.79 (p = 0.017) and a non-significant Wilcoxon test (p = 0.123, visit 1: 79 ± 14%, visit 2: 83 ± 14%). Between-experimenter reliability was tested in 158 dogs and resulted in a Cohen's kappa of 0.86 (p < 0.001). To assess attitudes to dogs, 126 members of shelter staff completed a questionnaire. Relationships between the approach behaviour of the dogs and the attitude of several subsamples of shelter staff (based on the time staff worked with dogs) were analysed on shelter level. In the subsample working 80% or more of their time with the dogs (N = 11), more dogs approaching the experimenter was related to a more positive attitude to dogs (rs = 0.68, p < 0.05) and a higher agreement to the use of positive dog handling (rs = 0.66, p < 0.05). However, more dogs approaching the experimenter was related to staff feeling less comfortable during interactions with dogs in almost all subsamples (rs ranging from −0.45 to −0.52, p < 0.05; N ranging from 19 to 28). We conclude that our approach test can reliably be used in a surveillance setting. With regard to validity, we found that a positive attitude and positive handling seem to increase the willingness of dogs to approach an unknown person. However, a high proportion of approaching dogs might also reflect a low frequency of human interactions, and therefore poor welfare, caused by staff feeling uncomfortable during interactions with dogs. Further research is needed to clarify how quantity of interactions with humans influences the approach behaviour of shelter dogs.