Summary A simple exercise in inverting transient electromagnetic (TEM) data for the layering in a basin ran into difficulties when the inversion failed to converge for about half of the soundings. The failure was caused by the forward modeling program that created artifacts for models with thin, shallow conductive layers. The artifacts were difficult to identify visually on raw plots because of the large dynamic range of most TEM data. They did show up clearly in relative comparisons of the curves from two programs. This experience led to a broader study of the differences between the many layered-earth and 3-D TEM modeling codes available and of the various approximations used. This study has already resulted in improvements in several codes, which generated more confidence in their results.