The prognosis in lung cancer depends largely on early stage detection, and thus new screening methods are attracting increasing attention. Canine scent detection has shown promising results in lung cancer detection, but there has only been one previous study that reproduces a screening-like situation. Here breath samples were collected from 122 patients at risk for lung cancer (smokers and ex-smokers); 29 of the subjects had confirmed diagnosis of lung cancer but had not yet been treated and 93 subjects had no signs or symptoms of lung cancer at the time of inclusion. The breath samples were presented to a trained sniffer dog squadron in a double-blind manner. A rigid scientific protocol was used with respect to earlier canine scent detection studies, with one difference: instead of offering one in five positive samples to the dogs, we offered a random number of positive samples (zero to five). The final positive and negative predictive values of 30.9% and 84.0%, respectively, were rather low compared to other studies. The results differed from those of previous studies, indicating that canine scent detection might not be as powerful as is looked for in real screening situations. One main reason for the rather poor performance in our setting might be the higher stress from the lack of positive responses for dogs and handlers.