Two professional tracking dogs were brought in at right angles to 50-m-long tracks to try to identify the clues that the dogs used to determine the direction of the tracks. When the dogs were tested on tracks laid by a walking man, they took the direction that the track-maker had moved significantly more often than predicted by random choice regardless of whether the person had moved forwards or backwards. When tested on tracks laid by bicycling on grass or asphalt, the choice of direction was not significantly different from random, even though the dogs were always rewarded when making correct choices. When four leather strips were tied to the back tyre of the bicycle before laying the track, the one dog tested took the correct direction significantly more often than predicted by random choice. When the leather strips were smeared with sausage before laying the tracks, the dog took the wrong direction more often than predicted by random choice. However, when rewarded for making correct choices, the dog learned to choose the direction in which the bicycle had actually moved. We suggest that our dogs determined the direction of tracks by comparing the scents of consecutive prints and following the prints in the direction of increasing intensity.