This big misconception comes from Raymond Chen's idea of E/R modeling.
That's the fundamental idea. But there's also a slightly more advanced understanding. That is that your tables/relations can't contain pointers. If you put pointers in your relations then you've created a graph/network database.
Think about encapsulating an array inside an object class and exposing "push" and "pop" methods. Wouldn't you agree that when you use an object of that class you're using a stack, not an array? Because the interface is what matters, not the implementation. The situation with pointers in your tables is slightly more complex, but roughly analogous.