We define an atom tree of a graph as a generalization of a clique tree: its nodes are the atoms obtained by clique minimal separator decomposition, and its edges correspond to the clique minimal separators of the graph. Given a graph GG, we compute an atom tree by using a clique tree of a minimal triangulation HH of GG. Computing an atom tree with such a clique tree as input can be done in O(min(nm,m+nf))O(min(nm,m+nf)), where ff is the number of fill edges added by the triangulation. When both a minimal triangulation and the clique minimal separators of GG are provided, we compute an atom tree of GG in O(m+f)O(m+f) time, which is in O(n2)O(n2) time. We give an O(nm)O(nm) time algorithm, based on MCS, which combines in a single pass the 3 steps involved in building an atom tree: computing a minimal triangulation, constructing a clique tree, and constructing the corresponding atom tree. Finally, we present a process which uses a traversal of a clique tree of a minimal triangulation to determine the clique minimal separators and build the corresponding atom tree in O(n(n+t))O(n(n+t)) time, where tt is the number of 2-pairs of HH (tt is at most View the MathML sourcem¯−f, where View the MathML sourcem¯ is the number of edges of the complement graph); to complete this, we also give an algorithm which computes a minimal triangulation in View the MathML sourceO(n(n+m¯)) time, thus providing an approach to compute the decomposition in View the MathML sourceO(n(n+m¯)) time.