The Nielsen–Schreier theorem states (in part):
Let $F$ be a free group, and $H\le F$ be any subgroup. Then $H$ is isomorphic to a free group.
I have seen the topological proof of this theorem using the correspondence between coverings and subgroups of the fundamental group. This has always struck me as using rather strong theory for what it is being used to prove (though I do appreciate the beauty of the argument).
In my head, I see the following (loose) argument:
Let $H$ be a subgroup of $F$, and assume that $H$ is not free. Then there exists some nontrivial relation $h_1h_2\dots h_n = 1$. But then this is also a nontrivial relation in $F$ implying that $F$ is not free, which is absurd. Thus $H$ is free.
Clearly, there must be some problem with this. What are the stumbling blocks here? An issue I see is that the exact notation a relation has always seemed a little vague to me (some reduced word equal to the identity?), but that doesn't seem like it ought to be a large enough problem to invalidate the argument.
This step is not obvious at all. To get a nontrivial relation in $F$, you need to know that when you expand $h_1,\dots,h_n$ in terms of the free generators of $F$, then $h_1\dots h_n$ reduces to a nontrivial reduced word. Why should that be true?
Indeed, it may not be true. You have glossed over the fact that you need to choose some specific subset of $H$ which will be your free generators. If you choose incorrectly, then there will be nontrivial relations. For instance, suppose $F$ is free on generators $a$ and $b$, and you take $H$ to be the subgroup generated by $h_1=ab$, $h_2=aba$, $h_3=bab$. Then $h_1^3h_3^{-1}h_2^{-1}=1$ is a nontrivial relation among these generators of $H$ (but when you expand this out in terms of the generators of $F$, it reduces to the trivial word, so this does not contradict freeness of $F$). So $H$ is not freely generated by $\{h_1,h_2,h_3\}$. To prove $H$ is freely generated, you have to somehow come up with a special set of generators for which there will be no nontrivial relations.
Now for this particular $H$, that is not so hard (in fact, $H$ is all of $F$, so you can take $a$ and $b$ as your free generators). But if you have some completely arbitrary subgroup of $F$, it is not at all obvious how you would come up with a generating set such that any relation between them would still be a nontrivial relation in terms of the free generators of $F$, thus giving a contradiction as you suggest.