This paper presents a perturbative model for the vibrational predissociation dynamics of inert gas hydrogen halide (RgHX) complexes. The predissociation is modeled as a Fermi Golden Rule (FGR) process from a bound state residing on a two-dimensional potential energy surface (PES) obtained by averaging over the HX vibrational state of interest to a series of one-dimensional exit channels obtained by averaging over the HX rovibrational state of interest. This model is applied to ArHF, for which a high-quality ab initio interaction potential is available. In particular, we focus on the upsilon --> upsilon - 1 transition for the bound states (1000), (2000), (2110), (3000), and (3110). We confirm the experimental observation that the product HF tends to come off at the highest accessible j state, which is j = 13 for this system. This results from a strong angular anisotropy in the ArHF interaction potential that couples low-j and high-j HF states. The basic mechanism for this high-j preference is determined to be the suppression of the low-j exit channels arising from highly oscillatory low-j outgoing wave functions. We also observed that the tails of the bound-state wave functions, in the inner wall region of the interaction potential, gave the main contribution to the predissociation rate, indicating that the vibrational predissociation process is due to tunneling and is therefore a purely quantum effect. The calculations also confirm the strong v-dependence of the predissociation rates, as well as the stabilization of the complex that occurs when energy is placed into the HF bending mode. For the (1000) and (2110) states, we obtain rates well below 1600 s(-1), which is consistent with the observation by Miller that the vibrational predissociation rates are too slow to be measured. The (2000) state does give a measurable rate, with a computed decay into the j = 13 exit channel of 14 000 s(-1). The (3000) state gives a corresponding rate of 200 000 s(-1), in good agreement with the overall dissociation rate of 250 000 s(-1). The (3110) rate is slower, with a value of 12 000 s(-1). Though this rate seems somewhat small, given that the lifetime of the (3110) state was measured to only be twice as long as that of the (3000) state, this rate and all our rates are within an order of magnitude of the measured rates.