A new short-term load forecasting procedure is presented in this work, mixing techniques from the statistical models and those from computational intelligence (CI). It takes advantage of the CI techniques to establish the various load profiles and to process climatic variables in a linguistic way, and those from the statistical side, where the parameters and the order of the model are known and a spread measure is determined. The model is an adaptation of the method developed by P.C.Gupta, where CI techniques are added to the original method. The final model includes a classifer scheme, a predictive scheme and a procedure to improve the estimations. The classifier is implemented via an artificial neural network using a non-supervised learning moving average, exponential smoothing and ARMA type of models. A fuzzy logic procedure uses climating variables to improve the forecast.