This paper presents a parametric modeling method for predesign cost estimation of urban railway systems. Data of 13 light rail and metro projects located in Turkey were compiled for quantification of the impacts of parameters on the project costs. Parametric models have been developed using regression analysis and neural networks techniques. Ten linear regression models were used for determination of the parameters significantly impacting cost of urban railway projects. Two neural networks were considered as an alternative to regression models, particularly for the identification of the non-linear relations. Predictive behaviour and performance of the models were compared to determine a model that presents adequate relations and has a reasonable accuracy. The proposed method provides a powerful approach for determination of a satisfactory parametric cost model during early project stages by incorporating a coordinated use of regression analysis and neural network techniques.