Rainfall-runoff process is one of the most important and complex phenomena in the hydrological cycle and therefore different views have been presented for modeling the phenomenon. Obviously, the recognition of the behavior of the catchment can play an important role in selecting the appropriate model as well as saving time on the simulation. Previous studies have shown that the multi-linear models have an acceptable performance in the case of watersheds which usually have a regular rainfall pattern. In this study, the multilinear Wavelet-M5 model was introduced and the rainfallrunoff process in the Aji Chay catchment was investigated. At first, the main rainfall and runoff time series were decomposed to several sub-time series by the wavelet transform to overcome its non-stationarity. Then the obtained sub-time series were imposed as input data to M5 model tree to forecast the runoff values and also the results were compared to the other models (i. e. ANN, M5 and WANN) by the root mean squared error and determination coefficient criteria. The results showed that the performance of the proposed hybrid Wavelet-M5 model increased up to 69% compared to the sole M5 model tree for the Aji Chay catchment.