Family dogs successfully follow human-given cues in a two-object choice test. However, whether this ability has any roots in dog–dog visual communication, has been seldom investigated. We designed a test where a video-projected, life-sized dog ‘demonstrator’ provided directional cues for the subjects by turning its head toward one of the two plates containing food. To avoid pseudo-replication, videos of 28 different dogs were recorded as ‘demonstrators’. N=60 subjects were tested with a larger, smaller, or equally sized ‘demonstrator’, using a momentary head-and-gaze turn, while N=15 subjects were tested using sustained demonstrative cues. Social status of the subjects, based on a questionnaire (single-kept, dominant or subordinate), was also taken into account. Our results showed that dogs did not choose between the plates by relying on the momentary head turns of the projected demonstrator (p=0.35). Social rank did not have any effect on their choices either, however, ‘single’ dogs showed a left-side preference in their choices (p=0.03). In the case of momentary gazing, all dogs developed strong side bias depending on their first choices (p<0.001). However, when the projected demonstrator performed sustained gazing, subjects avoided the indicated plate (p=0.027). This is the first study showing that dogs do not follow the head turn cues given by another dog, but rather avoid the food that the other dog gazes at. This finding gives support to the theory that good performance of dogs in case of human-given cues may not have a direct predecessor in dog–dog communication, but rather relies on dogs’ specific attention to humans and/or the plentiful opportunity to associate human directional gestures with food reward during the ontogeny.