Seismic ray tomography is a popular tool for reconstructing seismic velocity models from traveltime data. Here we study how the model parameterization affects the resolution and accuracy of the tomographic inversion for the near-surface model building. In particular, we consider the weighting of the elements of the model perturbation vector based on the values of the initial velocity model. When the model parameters are defined in terms of velocities, then the tomographic-inversion resolution is better for the shallow part but degrades for the deeper part of the model. The opposite is true when the model parameters are defined in terms of slowness values. This effect is associated with the method of forming the tomographic matrix. When linearizing the tomography problem for different model parameters, the matrix elements have different weight coefficients. This affects the inversion results and can lead to large errors. We suggest a new parameterization (in-between the velocity and the slowness) that provides better quality of the tomographic inversion and balanced resolution between the shallow and deeper part of the model. The good performance of this new parameterization is confirmed by a series of synthetic tests and one real-data example.