The dificulty of the Box and Jenkins approach for univariante time series forecasting lies in the stage of identification. The traditional methodology based on the estimators of the autocorrelation and partial autocorrelation functions, to mixed models(ARMA), usually leads to wrong structural identification. As an attempt to solve this problem, many authors have porposed alternative identification methods. This work intends to make a critical analysis was concentrated on the following methods: - Inverse Autocorrelation Function, (Cleveland, 1972) and (Chatfield, 1979); - R&S Arrays, (Gray, Kelley and Mc. Intire, 1978); - The Corner Method, (Beguin, Gourieroux and Monfort, 1980); - Extended Autocorrelation Function (Tiao and Tsay, 1982); - General Partial Autocorrelation Function (Glasbey, 1982); their performance were compared with each other and with the traditional method. Five structures have been studied: AR(1), AR(2), MA(1), MA(2) and ARMA(1,1). For each of them three models have been chosen /considering their position in the stationary and invertible regions. Fifteen series have been simulated for each model, varying levels and their seeds, adding up to 225 series, which were submitted to each method. The results led to several conclusions, which are restricted to the sample studied; the most important was: Although some of these methods yield to better results than the traditional ones, the problem of identification is still unsolved. So, any kind of Automatic Box and Jenkins can not be recommended.