Cooperative　distributed　system　covers　a　wide　range　of　applications　such　as　the　systems　for　industrial　controlling　and　business-to-business　trading,　which　are　usually　safety-critical.　Coordinated　exception　handling　(CEH)　refers　to　exception　handling　in　the　cooperative　distributed　systems,　where　exceptions　raised　on　a　peer　should　be　dealt　with　by　all　relevant　peers　in　a　consistent　manner.　Some　CEH　algorithms　have　been　proposed.　A　crucial　problem　in　using　these　algorithms　is　how　to　develop　the　peers　which　are　guaranteed　coherent　in　both　normal　execution　and　exceptional　execution.　Straightforward　testing　or　model　checking　is　very　expensive.　In　this　paper,　we　propose　an　effective　way　to　rigorously　develop　the　systems　with　correct　CEH　behavior.　Firstly,　we　formalize　the　CEH　algorithm　by　proposing　a　Peer　Process　Language　to　precisely　describe　the　distributed　systems　and　their　operational　semantics.　Then　we　dig　out　a　set　of　syntactic　conditions,　and　prove　its　sufficiency　for　system　coherence.　Finally,　we　propose　a　global-to-local　approach,　including　a　language　describing　the　distributed　systems　from　a　global　perspective　and　a　projection　algorithm,　for　developing　the　systems.　Given　a　well-formed　global　description,　a　set　of　peers　can　be　generated　automatically.　We　prove　the　system　composed　of　these　peers　satisfies　the　conditions,　that　is,　it　is　always　coherent　and　correct　for　CEH.