You can imagine the stack as being like a stack of plates. As soon as a function calls some other function things get added to the stack (from the bottom, upwards) starting with the return address (i.e. where the program flow should go back to, once the called function has finished).
Let's suppose there was nothing on the stack to begin with. One of the functions (either the calling function or the called function) would start a stack by placing a return address - and some extra 'plates' might be added, representing any arguments that get passed to the called function (the called function's parameters).
Program execution then gets passed to the called function which probably adds some more plates in the form of 'local variables'. Every 'plate' has some sort of information written on it that means something special to te program. By the time the called function reaches its return statement it has a big stack of plates. Just before it returns, it removes the plates (one by one) from the top downward so that it can read the return address, written on that very bottom plate. The program then jumps back to that original return address.
The most important thing about the stack is that you don't need to think much about it. Your program manages it. Also, stack management is very efficient because the plates don't literally need to be removed (like they would if they were real plates). In reality, the program contains a "stack pointer" that simply gets updated each time a particular plate needs to be written to or read from. The size of the stack is usually quite small because it's mostly used for local variables (variables that go out of scope pretty soon after they're finished with). The size of the stack is set at compile time and is small enough to be available on any computer, no matter how little RAM it's got.