In a quite informal way the axiom schema of replacement says that
Let $X$ be a set.
Let $\phi(x,y)$ a formula s.t. for each $x$ in $X$ there exists an unique set $y$ satisying $\phi$.
Then there exists the family $Y$ composed by all such $y$.
Therefore even if I know that any single set $y$ exists, without this axiom I cannot conclude that the family $Y$ exists.
In the finite case this should not be true, because from single sets $a$, $b$ and $c$ I can build $\{a,b,c\}$ by means the axiom of pair and the union axiom.
Therefore I suppose that we need this axiom schema just for the infinite case.
Am I wrog?
Thanks in advance.
That depends very much on your axiomatization of $\sf ZF$.
Specifically, the axiom of pairing is a consequence of replacement, power set, and infinity (one could use "empty set" instead of infinity, but that would be redundant, since infinity implies the empty set exists directly).
If minimality is what your heart desires, then pairing is a theorem, not an axiom, and then I don't see why things work out in the finite case either.
But let's just put this aside, and let pairing be part of our system. Indeed, then in that case you need the axiom of replacement for infinite collections.
The classic example is $V_{\omega+\omega}$. We start with $V_0=\varnothing$, and for $n$, $V_{n+1}=\mathcal P(V_n)$. When we reach $\omega$, $V_\omega=\bigcup\{V_n\mid n<\omega\}$, and so we continue again with power sets and unions.
It is not hard to check that $V_{\omega+\omega}$ actually satisfies all the axioms of $\sf ZF$ except Replacement. Including pairing, just to be clear. But now consider the function given by $f(0)=V_\omega$ and $f(n+1)=\mathcal P(f(n))=V_{\omega+n+1}$. The range of $f$ is exactly $\{V_{\omega+n}\mid n<\omega\}$, and it is easy to see that this is not an element of $V_{\omega+\omega}$.
So indeed Replacement fails, and since pairing and union hold, it fails for infinite sets.