This paper addresses the problem of predicting the express local train choices of metro passengers. The model was built and tested on the preferences observed from smart card data. The revealed preference data, because of intensiveness, can also accurately capture the marginal effects of the core attributes, in-vehicle time and wait time, on the express train choices by metro passengers. To be specific, the marginal disutility of a path decreases in in-vehicle time and increases in wait time. Accordingly, this paper employs a Box–Cox transform to adjust the constant marginality of a linear model to the nonconstant marginal disutility. The resulting nonlinear logit model improved the predictability of a conventional linear model. Tested on the Incheon–Yongsan interval of the Gyeong-In Line of the Seoul metropolitan area in South Korea, the model predicted a correct choice by a passenger in 99.9% and 99.5% of the cases during peak and nonpeak hour periods, respectively, compared with 96.9% and 95.8%, respectively, from a linear model. The model, applied to Line 9 without a parameter tuning, achieved a predictability greater than 95%.