A convenient and acceptable technique to develop mathematical models is through conceptual formulation and statistical development while integrating the effects of various variables on these physical processes.  Overemphasizing on these techniques could result in an increase in complexity of model which in turn influence the performance of the model. In this study, one conjunction model combining wavelet-neuro-fuzzy for runoff forecast is proposed and compared with simple neuro-fuzzy inference system. The inflow series to the conjunction model has been decomposed by wavelet transform. The performance of the conjunction model under the changed inflow parameters has been compared with the simple model. The results show that both the model performed well, however, increase in complexity of a model does not necessarily increase the performance of the model.