Cost Function Network (CFN) comes from Artificial intelligence fields. The approach has been recently applied with success for computation protein design. But, It can be used in various molecular modeling or design contexts to explore a large discrete energy landscape (the search space). CFN is a combinatorial optimization method. The aims are, if the chemical composition doesn't change, to minimize the total energy and found the most probable geometry. Otherwise, in the design context, the goal is to found a couple polymer sequence and geometry deriving from the same template with the best energy (i.e. maximum probability). The search provides solutions with optimality guaranty or not. From a theoretical point of view, it can be applied when the molecular energy can be decomposed as tailor Serie and consequently, precomputed the combinatorial problem (i.e. the energy landscape). Then, CFN exhaustively explored the search space with a branch and bound algorithm in order to find the optimal solution. The solution optimality can be proofed for Search space size including 10^234 stats, potentially more or less depending very likely on the energy landscape shape. After introducing the key concepts of cost function network, several Computational Protein Designs will be presented in order to illustrate the performance level and the limitations of toulbar2 our CFN solver. (http://www7.inra.fr/mia/T/toulbar2/)