This paper deals with forecasting of monthly sardines catches in north area of Chile. The forecasting model is based on un-decimated stationary wavelet transform (SWT) combined with radial basis function (RBF) neural network and linear autoregressive (AR) model. The original monthly sardines catches data are decomposed into two sub-series employing 1-level SWT and the appropriate subseries are used as inputs to the (RBF+AR) model to forecast 1-month ahead monthly sardines catches. The forecaster's parameters are estimated by using a hybrid algorithm based on the least square (LS) method and Levenberg Marquardt (LM) algorithm. The forecasting performance based on Hybrid (LS+LM) algorithm based was evaluated using determination coefficient and showed that a 99% of the explained variance was captured with a reduced parsimony and high accuracy.