Excessive use of fossil fuels, land-use change and the increasing world population and the ever-increasing of industrial activities accordingly to provide welfare and the needs of humans, thereby has been increased the concentration of greenhouse gases, particularly carbon dioxide in recent decades. This increase in greenhouse gases cause a phenomenon called climate change. The aim of this study was to evaluate the effect of climate change on surface runoff is Caesar's basin. In the present study first compares the performance of general circulation models (GCM) in simulation of climatic parameters includes temperature and precipitation averages. In the first step, performance of 6 Atmospheric general circulation models as: CGCM3, HADCM3, CSIROMK3 (collection of AR4 models) and CGCM1, GFDL30, NCARPCM (from the collection of model ATR) for simulation of climatic parameters includes average temperature and precipitation using artificial neural network (ANN) in Caesar basin were evaluated. For training of artificial neural network model was used perceptron forward method. According to performance evaluation of these models using maximum absolute error, mean absolute error , root-mean-square error and Russell Square Quality Representatives of the model, among AR4 and ATR models, AR4 models had better performance than ATR models and less uncertainty in simulating of climatic parameters (rainfall and average temperatures) in Caesar basin at 1996-2000. Between these 6 models, CGCM3 is the best performance in the simulation of climatic parameters for Caesar basin. CGCM3 and HADCM3 models have the lowest differences with observed climate parameters. Also, the results showed that CSIROMK 3.0 and CGCM1 models have the most differences with observed climate parameters. In the next step of this study, was used of HADCM3 model under A2 and B2 scenarios. For downscaling the SDSM model Ver. 4.2 was used. Based on the results of the SDSM calibration, between 26 large scale climatic parameters (NCEP) averagely 3 parameters have maximum correlation with the average temperature and 8 parameters with average rainfall in Caesar basin. SDSM calibration results in Caesar basin showed that RS (coefficient of explaining) were equal 0.86 and 0.26 for average temperature and rainfall respectively. These results indicate that SDSM model has necessary efficiency for the downscaling climatic parameters. The 1971-1996 period and 1995-2001 period, were used for calibration and accuracy assessment respectively. Results of validation based on Russell Square Quality Representatives of indicated HADCM3 model has a good performance in simulation of climatic parameters of Caesar. Climatic parameters for 2050-2021 periods simulated and compared with 1979-2008. The results indicate an increase in temperature and decrease in rainfall. To simulating of runoff procedure provided by Jakeman and Hornberger (1993) (model IHACRES) was used. The input data to this model are including temperature, precipitation and daily discharge. For calibration of rainfall-runoff model, temperature, precipitation and discharge data in daily scale entered to IHACRES model from 1975 to 1995. For validation of IHACRES model were used mentioned data above in 1996-2004 period. After evaluating of model performance, daily discharge was simulated for 2021-2050. The results showed in all the stations simulated daily discharge reduced Compared to observed daily discharge. The greatest difference between simulated discharge in two future periods in A2 and B2 scenarios compare to base period (2004-1975) in the Cham zaman station were 29 and 20 percent respectively. In fact, based on the simulated A2 and B2 scenarios, daily discharge deceased 29 and 20 percent than to the base period. This difference in Biaton station for A2 and B2 scenarios is 19 and 13% and in Rahim Abad station for A2 and B2 scenarios, to 24 and 15 percent respectively. Keywords: climate change, artificial neural network, model prspetron, model IHACRES, model SDSM, surface runoff, Caesar's watershed