The surface wave full ray theory (FRT) is an efficient tool to calculate synthetic waveforms of surface waves. It combines the concept of local modes with exact ray tracing as a function of frequency, providing a more complete description of surface wave propagation than the widely used great circle approximation (GCA). The purpose of this study is to evaluate the ability of the FRT approach to model teleseismic long-period surface waveforms (T ∼ 45–150 s) in the context of current 3-D Earth models to empirically assess its validity domain and its scope for future studies in seismic tomography. To achieve this goal, we compute vertical and horizontal component fundamental mode synthetic Rayleigh waveforms using the FRT, which are compared with calculations using the highly accurate spectral element method. We use 13 global earth models including 3-D crustal and mantle structure, which are derived by successively varying the strength and lengthscale of heterogeneity in current tomographic models. For completeness, GCA waveforms are also compared with the spectral element method. We find that the FRT accurately predicts the phase and amplitude of long-period Rayleigh waves (T ∼ 45–150 s) for almost all the models considered, with errors in the modelling of the phase (amplitude) of Rayleigh waves being smaller than 5 per cent (10 per cent) in most cases. The largest errors in phase and amplitude are observed for T ∼ 45 s and for the three roughest earth models considered that exhibit shear wave anomalies of up to ∼20 per cent, which is much larger than in current global tomographic models. In addition, we find that overall the GCA does not predict Rayleigh wave amplitudes well, except for the longest wave periods (T ∼ 150 s) and the smoothest models considered. Although the GCA accurately predicts Rayleigh wave phase for current earth models such as S20RTS and S40RTS, FRT's phase errors are smaller, notably for the shortest wave periods considered (T ∼ 45 s and T ∼ 60 s). This suggests that the FRT approach is a useful means to build the next generation of elastic and anelastic surface wave tomography models. Finally, we observe a clear correlation between the FRT amplitude and phase errors and the roughness of the models. This allows us to quantify the limits of validity of the FRT in terms of model roughness thresholds, which can serve as useful guides in future seismic tomographic studies.