Time series analysis is one of the essential and complicated research. It is a well-known fact that improving time series prediction accuracy is a vital yet often challenging issue. Recently, soft-computing has become popular in time series forecasting in various application areas. Soft-computing is a fusion of research of evo-lutionary algorithms and genetic programming, neural networks, fuzzy set theory, and fuzzy systems, and provides rapid dissemination of results. This study investigates a model for time series water-level prediction using soft-computing techniques. The aim of this study to develop and implement a model that is more relia-ble and effective. In study twelve (12) years of data of the Jamuna river were collected from the Bangladesh Water Development Board (BWDB). Besides,to evaluate the proposed model applied to the four areas (sta-tions): Aricha, Bahdurabad, Shariacandi, and Sirajganj of Jamuna river. In experiments, past 2 to 4 days’ time series data with rainfall and without rainfall has been applied to predict 1 to 4 days ahead water-level. The experimental results demonstrated that ANFIS performs superiorly to traditional methods, such as NARX and FTDNN.