Abstarct: Determining the load of suspended sediment in rivers is of most important factor in water engineering, hydraulics, and environmental; so, it has a strong effect on the design and management of hydraulic structures. The loss of agricultural land, filling reservoirs, and irrigation canals are the problems resulting from erosion and sedimentation. Continual measurement of flow discharge is difficult and time-consuming. It is more difficult even in flooding. Hence, by establishing stage-discharge relation, flow discharge of river can be estimated. So far, much attempt has been made to predict the suspended sediment load accurately, most of which were performed through establishing the relationship between sediment discharge and flow discharge to estimate sediment discharge. However, these regression conventional methods including sediment rating curve and FAO lack certainty. Therefore, many researchers have turned to intelligent methods such as artificial neural network, neuro fuzzy-comparative inference system, and evolutionary algorithms. In the present study, to predict suspended sediment discharge and flow discharge of Kore Sang's hydrometric station in Amol located on Haraz River, gene exxpression programming (GEP) and MARS methods were developed, and the obtained results were compared with that of sediment rating curve, FAO, and linear regression methods. For models input, flow discharge, water level, and station's sediment discharge data related to 1964-2013 were collected and examined. This study was conducted in two parts: In the first part, the data for flow discharge, water level, and sediment discharge was introduced as input into the model to predict sediment discharge. In the second part, the data for flow discharge and water level was introduced as input into the model, and a relationship has been established between flow discharge and water level. Comparing the results obtained and examining statistical parameters, including correlation coefficient (R2), root mean square error (RMSE), and mean average error (MAE) revealed that comparing other methods, GEP and MARS ones had a significant advantage in predicting. The study results also showed that MARS method had better performance in the first and second part of training phase, and GEP method had better performance in the last part of testing phase.