In this article flow prediction using ARIMA model has been done and the accuracy of ARIMA model in long-term forecasting has been tested. In this study, using yearly data (since 1995–2015), obtained from Hydrometric station at Hirakud (upstream of Hirakud Dam), the Auto Regressive Integrated Moving average (ARIMA) model is used for prediction of monthly mean inflow and daily inflow to Hirakud Dam reservoir. The XLSTAT, STATA and Microsoft Excel some of the software's which were used to model ARIMA and to validate the results. The methods adopted is to predict the forecasting of the run-off in an one year short-term basis and subsequently the predicted data of that year is included as an observed data set for short-term forecasting of run off for the next year and that value again is treated as observed data base. By this process when both monthly average and daily run off for the year 2030 is predicted when the observed data set used for its prediction is 1995 to 2029, though effectively the observed data set actually available is 1995 to 2015. On the basic of comparison of the results of the various candidate models with observed data like ARIMA (1, 1, 0), ARIMA (2, 1, 0), ARIMA (4, 1, 0) and ARIMA (5, 1, 0), the performance of (5, 1, 0) model is found to be acceptable for monthly stream-flow prediction as it gives comparatively more accurate result.