自1970年初期以來，多項Logit模型（簡稱MNL）已被廣泛地運用在於運具選擇、區位選擇以及住宅選擇等研究。然而，在模型之效用函數的參數指定方面，卻出現了不同指定方式的爭議。第一種方式是將參數指定成為替選方案成對比較之間的相對效用。因此，當替選方案的個數為n時，將會有n-1組之參數估計值，本研究稱此為原始多項Logit模型。另一種方式則是將參數指定成為與替選方案無關。無論替選方案之個數有多少個，則校估之參數僅有一組，本研究稱為一般多項Logit模型。兩者在基本假設的認定、參數校估的方式（包括電腦應用軟體）以及參數的解釋皆有所差異，而在實證應用上亦各有優缺點及限制。在1995年第十四屆太平洋區域科學會議中，此問題引起相當的爭議。本文擬就兩種指定方式的異同進行比較，從模型的假設開始，進行推論並相互比較。研究結果指出兩者就代數的角度而言，是可以相互推導，而在二項選擇之參數校估結果亦有高度相似性。此外原始多項Logit模型較適用於現況的解釋，而一般多項Logit模型則較適用於新替選方案被選擇機率的預測。

　　　　　Since early 1970's, the multinomial logit model has been used in the decision making such as mode choice, location choice housing choice and so on. However, there exists an debate regarding the specification of parameters in the utility function. One approach is to define the parameters to be alternative-dependent. Therefore the number of vector of parameters equals the number of alternatives subtracting by one. This approach is refered as the original multinomial logit (OMNL) model in this study. Another approach is to define the parameters to be the common marginal effects of variables. Hence, there is always only one set of estimated parameters no matter how many alternatives are in the choice set. We refered this approach as the general multinomial logit (GMNL) model. In this study, we compare these two models from the theoretical fundation, calibration problems and model interpretation. It can be concluded that these two models are identical in binary choice. The OMNL model is theoretically better in current behavior interpretation and the GMNL is favor of estimating the probability of new alternatives.