Accurate streamflow prediction is required in sustainable water resources management. Direct use of observed data in developing prediction models has resulted in inaccuracies. Discrete wavelet transform (DWT) is widely used to decompose observed data (raw data) into spectral bands and eliminate trends and periodicity to improve the accuracy of the models. However, DWT is known to have serious drawbacks, and predictions of daily streamflow have been with short lead times. In this study, a simple method called the SEASON algorithm was used to decompose the observed data into components with the objective of overcoming the drawbacks of DWT so that daily streamflow can be predicted with better accuracy and longer lead times. Data decomposed by SEASON and DWT were used as input into multilayer perceptron (MP) approaches to develop new approaches for predicting daily streamflow for lead times up to 7 days, and termed as seasonally adjusted series-multilayer perceptron (SAS-MP) and wavelet-multilayer perceptron (W-MP), respectively. Twelve years of approved daily streamflow data were obtained from Station 02231000 (located in the St. Marys River watershed) and Station 07288280 (located in the Big Sunflower River watershed), USA. Seven years of data were used for calibration (training) and the remaining 5 years of data were used for prediction (testing). The new approaches were compared with the stand-alone MP model by taking root mean squared error, coefficient of efficiency, and skill score into consideration. The results showed that the SAS-MP and W-MP models performed better than the stand-alone MP model, and the prediction accuracy increased with the use of decomposed signals. However, for all lead times, the SAS-MP model outperformed the W-MP model, which performed less after a lead time of 4 days. This indicates that the SEASON algorithm has the capability to capture periodicity better than DWT and can be used to extend lead time with better prediction reliability. (C) 2015 American Society of Civil Engineers.