The Gaussian beam method (GBM) has found wide application in seismology for charting high‐frequency wave propagation through heterogeneous environments. Unlike ray fields, Gaussian beams negotiate convergence zones without requiring caustic corrections. Because of these attributes, GBM may be useful also for sound field propagation in a realistically range‐ and depth‐dependent ocean. However, in its conventional and facile implementation via a discretized stack of paraxially approximated beams, GBM is not a discipline witha prioripredictive capability. By comparison with various canonical test problems, the accuracy of GBM has been found to depend strongly on the choice of arbitrarily assignable beam and stacking parameters. Here, the method is tested on previously not investigated multiply reflected fields in a range‐independent duct bounded by a rigid bottom. The fields on the boundary are computed by asymptotic ray theory (ART), GBM, and by a rigorous generalized ray integral whose numerical evaluation serves as a reference. It is found that a beam stack “tuned” for accuracy along a ray that has undergone a given number of reflections becomes successively less accurate as the observer moves to a ray segment with a lesser or greater number of reflections. Retuning removes this deficiency but requires again a reference result for comparison. This circumstance illustrates again the basic difficulty with GBM. Possible spectral remedies [Lu, Felsen, and Ruan, Geophys. J. R. Astron. Soc. (1987)] are discussed, but they are less readily computable. [Work supported by NJIT and ONR.]