We define a class of tree schemata and a notion of refinement between trees (finite or infinite). These concepts allow us to talk about different "ways to program an algorithm" and the "structure" of an algorithm. We obtain as a special case certain concepts such as "approximation", "convergence" and "fixed point semantics" recently employed by Scott to describe the semantics of flow diagrams. We concentrate in this paper on relating our concepts to Scott's.