Suspended sediment (SS) flux in a river is an important parameter for the watershed management. The large extend of watersheds areas and limited of sediment stations measurement have been caused that different methods have been developed to SS estimation. In this study, 42 sediment measurement stations existed in south and southeast of the Caspian Sea with over 20 year period was chosen. The sediment rating curve (SRC) was drawled using discharge and SS load measurements for 42 stations, then daily SS load was estimated by using the daily discharge (Q) and the average of annual SS was computed. By using the gamma test (GT), 14 effective variables on SS was reduced to 5 main variables and the mentioned stations were recognized in four homogeneous groups by using Andrew c urves (AC). For each homogeneous group and all existed stations using five selected main variables was developed regression models for average of annual SS load. Relative error (RE), r elative bias ( RBIAS) and r oot relative mean squared error ( RRMSE) indices statistics are used for evaluating the accuracy of the models. The results indicated that a model was not provided for group 1 . The models of g roups 2 to 4 have a good accuracy and the model of g roup 4 with RE(29%), RBIAS(12%) and RRMSE(35%) in the calibration stage , and RE(19%), RBIAS(12%) and RRMSE(25%) in the validation stage has the best performance and One model for all stations has high error . Peak discharge with two return periods (Qp) has the most impact on the average of annual SS change.