We present a theory of mode-specific tunneling that makes use of the general tunneling path along the imaginary-frequency normal mode of the saddle point, Qim, and the associated relaxed potential, V(Qim) [Y. Wang and J. M. Bowman, J. Chem. Phys. 129, 121103 (2008)]. The novel aspect of the theory is the projection of the normal modes of a minimum onto the Qim path and the determination of turning points on V(Qim). From that projection, the change in tunneling upon mode excitation can be calculated. If the projection is zero, no enhancement of tunneling is predicted. In that case vibrationally adiabatic (VA) theory could apply. However, if the projection is large then VA theory is not applicable. The approach is applied to mode-specific tunneling in full-dimensional malonaldehyde, using an accurate full-dimensional potential energy surface. Results are in semi-quantitative agreement with experiment for modes that show large enhancement of the tunneling, relative to the ground state tunneling splitting. For the six out-of-plane modes, which have zero projection on the planar Qim path, VA theory does apply, and results from that theory agree qualitatively and even semi-quantitatively with experiment. We also verify the failure of simple VA theory for modes that show large enhancement of tunneling.