Using Abrasive waterjet (AWJ) for rail treatment was first proposed by our research team. In order to further clarify the parametric response of AWJ on kerf characteristics including the top kerf width (ω), the kerf taper (θ) and the kerf depth (h), experiments were carried out on a section of the rail specimen in this study. Experiments were designed using the Box Behnken Design (BBD) method with three factors, which are waterjet pressure (P), traverse speed (TS) and standoff distance (SOD). The regression models were established to predict the kerf quality by using response surface methodology (RSM), and experimental verification was carried out. The results show that ω is most obviously affected by the SOD, while θ and h are most sensitive to TS. The predicted value of the model is basically consistent with the experimental value of the verification. The maximum error is less than 9%, and the average error is less than 6%, indicating that the established model can effectively predict the kerf characteristics of AWJ cutting rails.