A roughness scaling behaviour is tested by performing the direct numerical simulation (DNS) of a turbulent channel flow over three-dimensional sinusoidal rough walls. By systematically varying the roughness height ${{k}^{+}}$ and the roughness steepness S, the results for three groups of cases are considered and compared with those for flat-wall turbulence. The results show that the mean velocity and Reynolds stresses are highly dependent on both ${{k}^{+}}$ and S. To describe these specific relationships, we define a coupling scale ${{k}^{+}} S$. With this coupling scale, all the simulated data for the roughness function (${\rm \Delta} {{U}^{+}}$), the ratio of the pressure drag to the total wall resistance (${{\gamma }_{p}}$), the normalized bulk mean velocity ($U_{b}^{+}$) and the peak of the streamwise turbulent velocity fluctuations ($\overline {u_{p}^{\prime +}}$) collapse onto single curves, which shows that there is a strong direct correlation between them, i.e. ${\rm \Delta} U^{+}, \gamma _{p}, U_{b}^{+}, \overline {u_{p}^{\prime +}} \propto f(k^{+} S)$. Furthermore, a model for the prediction of wall resistance based on the roughness function can be established by defining a drag increasing ratio (DI). Accordingly, the wall resistance coefficient ${{C}_{f}}$ can be estimated directly from ${{k}^{+}}S$ of a given rough surface. These results suggest that this coupling scale provides a useful alternative to the equivalent sand grain roughness ${{k}_{s}}$.