We introduce a method for the construction of skeletal curves from an unorganized collection of scattered data points lying on a surface. These curves may have a tree like structure to capture branching shapes such as blood vessels. The skeletal curves can be used for different applications ranging from surface reconstruction to object recognition. As an input, the algorithm takes a set of 3D points. It returns a set of curves arranged in a tree structure. The only interaction needed is the selection of a data point which represent the root of the tree. A neighborhood graph is constructed over the set of points to compute geodesic distances between the root point and the other points. Connected level sets of the distance map are then extracted and organized in a tree structure. The centers of these levels sets constitute the skeletal curves.