Among all available forecasting techniques, Box-Jenkins technique is one of the most powerful and accurate forecasting techniques known today. Despite its accuracy, the use of Box-Jenkins technique is still very limited due to the high level of knowledge required in comprehending the technique and to the cumbersome iterative procedure which requires a large amount of cost and time in applying the technique to the real data. A rather direct way of overcoming this limitation and thus enhancing the wide use of Box-Jenkins technique is to automate its modelling procedure. This thesis proposes a method of automating the univariate Box-Jenkins modelling procedure by using Variate Difference method, D-statistic and Pattern recognition algorithm combined with Akaike``s Information Criterion. The results of the application to real data show that the average performance of automatic modelling procedure is better or not worse, at least, than those of the existing models modeled by specialists manually.