現代企業資料庫中,往往擁有龐大的顧客交易資料以及顧客基本資料。為了從資料庫中瞭解顧客行為,目前業界最常使用的是簡單捕捉顧客行為模式的RFM模型,並利用RFM模型找出對企業貢獻度高的顧客,進而進行顧客關係管理。本研究以RFM模型為基礎,將消費頻率的分布情形與變動權重的概念導入,引用「連」(Run)的概念,將一般評估顧客指標的RFM模型加以修正。本文以廣義伽瑪分配模擬出顧客的交易時間,透過線性判別分析比較Hughes和Stone指標,以及本研究所提出的新指標法(S),這三種RFM模型的顧客判斷正確率以及判斷錯誤所造成的誤判成本。本文研究結果發現,我們提出的新指標法(S),在重要顧客的判斷正確率上明顯穩定且優於其他兩者;在誤判成本的部份亦低於其他兩種評估方法。 In database of modern enterprise, there are huge amont of customer trade materials and personal information. In order to analyze customer behavior from the database, the RFM model which catches the patterns of customer behaviour is used frequently to find out important customers and to perform customer relationship management. This research proposes a modified RFM index based on the distribution of the consuming frequency, variable weights, and the concept of "Run". With simulated customer''s trading time by the generalized gamma distribution, we compare the performance of Hughes indicator, Stone indicator, and our new S indicator. We find that our new S indicator is more stable and performs better in the accuracy of the determination of important customers, and has lower misclassification costs.