Conceptual and mathematical models are presented that explain tracer breakthrough tailing in the absence of significant matrix diffusion. Model predictions are compared to field results from radially convergent, weak-dipole, and push-pull tracer experiments conducted in a saturated crystalline bedrock. The models are based upon the assumption that flow is highly channelized, that the mass of tracer in a channel is proportional to the cube of the mean channel aperture, and the mean transport time in the channel is related to the square of the mean channel aperture. These models predict the consistent -2 straight line power law slope observed in breakthrough from radially convergent and weak-dipole tracer experiments and the variable straight line power law slope observed in push-pull tracer experiments with varying injection volumes. The power law breakthrough slope is predicted in the absence of matrix diffusion. A comparison of tracer experiments in which the flow field was reversed to those in which it was not indicates that the apparent dispersion in the breakthrough curve is partially reversible. We hypothesize that the observed breakthrough tailing is due to a combination of local hydrodynamic dispersion, which always increases in the direction of fluid velocity, and heterogeneous advection, which is partially reversed when the flow field is reversed. In spite of our attempt to account for heterogeneous advection using a multipath approach, a much smaller estimate of hydrodynamic dispersivity was obtained from push-pull experiments than from radially convergent or weak dipole experiments. These results suggest that although we can explain breakthrough tailing as an advective phenomenon, we cannot ignore the relationship between hydrodynamic dispersion and flow field geometry at this site. The design of the tracer experiment can severely impact the estimation of hydrodynamic dispersion and matrix diffusion in highly heterogeneous geologic media.