本研究利用分析網路程序法 (Analytic Network Process, ANP)，透過擷取專家經驗與判斷，進行台灣地區印表機營業量之預測。期望在難以獲得大量歷史資料且涉及社會層面廣泛時，利用關係人的意見以及兩兩成對比較之一模糊方式，建構ANP預測模型。其後再進行敏感度分析以驗證本模型之穩健性後，並與環保署基管會現行統計方法進行比較，以確認本法之適用。 受全球氣候及生態變遷之影響，各國無不將環保議題列為關注焦點。而台灣為執行資源回收之先驅國家，自民國60年起即進行資源回收與廢棄物清除之相關作業，並訂法律規範使其具有完整約束效力。於現今資訊化時代，印表機已成為個人及企業不可獲缺之資訊輸出品。而近年來對印表機的需求、使用行為趨於複雜化，造成廢印表機大量增加及難以進行推估。 民國90年基管會增列回收廢棄印表機，透過向責任業者徵收清除處理費，作為補貼末端執行回收處理業者之費用，而營業量及廢棄量皆為費率考量的要素因子之一。其預測準確度將會影響此政策執行的成效。 進行傳統預測分析時受限於需要多筆歷史資料，且難以反應複雜因素的相依關係與其即時變化現象。因此透過ANP將能改善及克服傳統預測方法所遭遇之困境，並建立一具有彈性且良好效果之預測模型，以利環保工作之執行。 This research apply Analysis Network Process (ANP) to collect experts’ judgments on forecasting sales volume of printer in Taiwan through pairwise comparison. When lacking of history data and clarity of social impacts, ANP technique can be constructed for forecasting. A sensitive analysis is also made to assure the forecasting model to be robust. Because the global warming and our ecological system are changed seriously, our concerns are forced to these issues. In 1971, Environmental Protection Administration (EPA) of R.O.C. government had started some actions on waste clean-up and resources recycling and established some regulations to improve recycling rate to decrease environmental deterioration. The used printers belong to one major part among electronic wastes, and the amount of printers to selling and recycling are difficult to estimate due to the complicated customers’ behavior and economic situation. In the year of 2001, Recycling Fund Management Board (RFMB) of EPA initiates the action of recycling used printers in Taiwan. The Board collects funds from manufacturers and importers when they sell or import printers, and subsidize recycling industries with recycling and treatment fee to increase recycling ratio. Sales volume and waste collected volume are the major factors for setting up the fee, and both are relied on forecasting. Therefore, the accuracy of forecasting has a great impact on the performance of recycling. There are a couple of limitations in conducting traditional forecasting tools, most of them are statistical methods. The first one is that the methods need sufficient amount of data, which might be impossible sometime. The second one is that they are hard to response the real-time moves. Processing dependence among concerned factors of the real world is the most unfavorable limitation. Hence, we choose ANP to divert the limitations, and establish a forecasting model with good performance and a flexible structure for recycling printers.