Identification　of　the　non-linear　joint　parameters　of　the　complex　structure　with　force-state　mapping　in　frequency　domain,　the　number　of　the　data　required　in　the　curve　fitting　can　be　drastically　reduced　compared　to　the　time　domain　approach.　Because　the　model　of　the　joint　is　not　known　in　advance,　and　noise　affected　the　identification　accuracy　cannot　be　ignored,　when　using　least　square　method　to　solve　the　over-determined　equation,　due　to　the　inverse　operation　of　the　matrix,　the　noise　may　have　great　effect　on　the　identified　parameters,　the　equation　will　be　ill-condition　and　the　ill-posed　problem　arise　from　the　noise　effect.　A　novel　method　to　identify　the　nonlinear　joint　parameters　is　proposed　from　two　aspects　as　follow:　the　assumed　model　includes　all　the　exact　terms　and　some　surplus　term;　the　assumed　model　does　not　include　all　the　exact　terms　and　has　some　surplus　terms.　A　model　combining　Tikhonov　regularization　and　iterative　method　is　suggested　to　update　the　joint　model.　It　is　found　that　even　if　the　assumed　model　does　not　include　all　the　exact　terms,　as　long　as　model　updated　continually,　meet　the　convergence　criteria,　the　identification　accuracy　is　still　high.　Example　analysis　confirms　that　this　method　has　high　identification　accuracy.　©　2011　Journal　of　Mechanical　Engineering.