Little is known about whether or not cows can learn to use a visual cue to navigate complex paths. Experiment 1 investigated discrimination learning of dairy cows where the discrimination problems involved selecting a signalled one-way pair of gates in cattle yards. All cows learned to select the pair of gates signalled by a yellow sign (S+) to gain access to food. Learning to approach the yellow sign did not transfer to a new situation and problem. In Experiment 2, four cows were trained to select the correct path in a T-maze and then to reverse location. The cows with the yellow sign signaling the correct arm were more accurate at reversing responses than cows without the yellow sign. Reversal learning transferred to new T-mazes presented within the same yard setting. In Experiment 3, two heifers were provided with yellow signs and two heifers were not in a double T-maze with no prior reversal learning experience. There was no difference in performance shown by the sign and no-sign heifers, suggesting that maze learning was not being aided by the use of signs. Experiment 4 used the same stimuli as in Experiment 3 and trained cows in a reversal task before presenting a double T-maze and then eight complex maze problems with and without signs. The data suggest that cows learned to solve all mazes presented, and were better at selecting the correct path in maze problems with signs present. Experiment 5 also used a reversal task to train cows to select the arm of a T-maze when it was signalled by a yellow sign. However, transfer of learning was not shown when cows were tested in a new setting. These experiments show that cows can learn to use signs to select the correct path with reversal learning to solve maze problems, but in order to transfer such learning to a new setting, cows may need to have stimuli common to the training setting also present in any new setting.