A general formalism is developed for describing the effects of restricted rotational diffusion on deuteron ( H-2) magic angle spinning (MAS) nuclear magnetic resonance (NMR) spectra. The approach is based on the Smoluchowski model that describes restricted rotational diffusion in an arbitrary ordering potential with an arbitrary diffusion tensor. It is shown that the Smoluchowski model gives a physically more reasonable description of molecular motion than the discrete Markov (jump) model. The models are shown to be mutually consistent for high ordering potentials and (or) low temperatures provided the diffusion coefficient is sufficiently high. However, for low ordering potentials and (or) high temperatures the discrete Markov model is not a useful approximation and the spectra can only be simulated with restricted rotational diffusion. This is also the case for small diffusion coefficients independent of the ordering potential and the temperature. The formalism is based on finite difference solutions to the stochastic Liouville-von Neumann equation. This defines a linear homogeneous system of coupled parabolic partial differential equations which includes both first- and second-order spatial derivatives. Numerical solutions are very difficult to obtain and some useful finite difference methods are presented. The results are elaborated for H-2 MAS NMR spectroscopy. Solutions are obtained both in the presence and absence of radio frequency (rf)irradiation and effects of finite pulse width are included. The method is applied to the investigation of motional effects on H-2 MAS NMR spectra of L-alanine-N,N,N-H-2(3). The orientational dependence of the ordering potential and the quadrupole parameters is determined from the Smoluchowski model. The activation energies are found to be temperature dependent. These effects have not previously been observed and give new information on molecular motion in this system. The rotational diffusion results are compared with the discrete Markov model and it is found that in this case the two models are consistent. The most important difference is that the Markov model does not map out the orientational dependence of the ordering potential and the quadrupole parameters. Another advantage of the rotational diffusion model is that it is physically more reasonable than the Markov model and that the parameters may be interpreted in terms of molecular properties.