A　water　balance　model　A　water　balance　model　is　developed　for　arid　regions　in　this　study.　The　partition　calibration　strategy　is　also　proposed　for　calibrating　model　parameters　such　as　river　loss　coefficients　and　irrigation　return　water　coefficients　by　adopting　the　method　of　the　modified　dynamically　dimensioned　search　algorithm　(MDDS).　The　conditional　probability　and　Bayesian　statistics　are　employed　to　demonstrate　the　theoretical　rationality　of　partition　calibration　strategy.　The　case　study　in　the　Kaidu　River　Basin　of　Xinjiang　Uyghur　Autonomous　Region　has　verified　it　in　its　application.　According　to　the　model　results　of　Yanqi　and　Baolangsumu　hydrological　stations,　the　partition　calibration　strategy　is　superior　to　the　whole　regional　calibration　in　terms　of　parameters　number,　computing　time　and　Nash　efficient　coefficients.　Therefore,　the　partition　calibration　strategy　is　suggested　to　be　applied　to　these　areas　with　the　obvious　advantages　of　sub-partition　computing　units　and　parameter　distribution.　©　2014　IEEE.