The study is based on box theory by Nicolas Darvas and golden spiral of measurement by the recurrent trend of the stock to analyze the stock of upside and downside repeatedly. Using the box theory, golden spiral of measurement and multiple indexes to establish the knowledge base of fuzzy rules, we construct a three layers back-propagation neural network to integrate the knowledge base of fuzzy rules. When completing training fuzzy back-propagation neural network, we are capable of forecasting the price fluctuation and trend of the stock.