When you represent an unrooted tree in Newick format, an arbitrary node is chosen as the root. Thus, you can have different Newick forms for the same tree depending on where this root is placed. Also, you can move the order of nodes around without changing the topology, e.g. ((A,B),(C,D)) and ((C,D),(A,B)) represent the same tree.

A description of the Newick tree format is given here, on Joe Felsenstein/Mary Kuhner's lab webpages

http://evolution.genetics.washington.edu/phylip/newicktree.html

As Farhat says above, Newick represents a rooted tree. Convention is to represent a tree that is binary and unrooted with a polytomy/multifurcation at the root node. However, note that this is only convention. The tree (A,B,(C,D)) may represent a binary, unrooted, four-taxon tree. But it might also represent a non-binary rooted tree with a polytomy at the root, the root node linked to two terminal branches (one leading to the OTU A, the other to OTU B), and an internal branch leading to the internal node that is linked to the external branches leading to OUTs C and D (OTU: Operational Taxonomic Unit http://en.wikipedia.org/wiki/Operational_taxonomic_unit )

Note the cute "paradox" that most methods of tree estimation used these days estimate unrooted trees, but for many applications of trees, we want/need to make some inference about where the root is (or isn't). Hence we need to find some way/assumptions about the position of the root (ideally that we can defend, and that hopefully we state when presenting our rooted trees) to be able to use the trees we estimate for applications requiring rooted trees.