Abstract. We present some techniques for handling planning problems with numerical expressions that can be specified using the standard planning language PDDL. These techniques are implemented in LPG, a fully-automated planner based on local search that was awarded at the third international planning competition (2002). First, we present a plan representation for handling numerical expressions called Numerical Action Graph (NA-graph). Then, we propose some extensions of LPG to guide a search process where the search states are NA-graphs. Finally, we give some experimental results showing that our techniques are very effective in terms of CPU-time or plan quality, and that they significantly improve the previous version of the planner.