In W. W. Rouse Ball's Mathematical Recreations and Essays (11th edition 1939, and probably earlier editions) states (p.181): "Warnsdorff [sic] added that when, by the rule, two or more cells are open to the knight, it may be moved to either or any of them indifferently. This is not so, and with great ingenuity two or three cases of failure have been constructed, but it would require exceptionally bad luck to happen accidentally on such a route." However he gives no reference to where work showing this was done, and diagrams no example of it. The diagrams shown here are the first cases I have encountered where the rule fails.
The choices of route available from f7 onwards all lead to a dead end, shown by the darker circle, that leaves two cells, marked by crosses, unvisited.