A scale-dependent dynamic Smagorinsky (SDDS) model proposed by Porte-Agel et al. (2000) was introduced into the large-eddy simulation (LES) of a turbulent flow over a two-dimensional steep hill in order to improve the poor prediction accuracy of the scale-invariant dynamic Smagorinsky (DS) model reported in our previous paper (2004). The prediction accuracy was assessed by comparing the results with those from an experiment by Ishihara et al. (1998), the standard Smagorinsky model, and the DS model. On the whole, the SDDS model gives accurate predictions and improves the poor prediction accuracy of the DS model; however, the improvement is somewhat insufficient. Furthermore, the computational cost of the SDDS model was much higher than that of the other two SGS models, mainly due to the cost of solving the polynomials for the scale-dependence parameter.