Abstract.A statistical analysis of the properties of the plasma sheet in the geomagnetic tail using the ISEE 3 deep-tail data set is undertaken. The observed plasma sheet properties are compared to predictions of two models of reconnection.The first model [Cowley, 1980;Cowley and Southwood, 1980] is based on the balance of stress between a cold plasma inflow and the magnetic tension of reconnected field lines at the current sheet. The second [Owen et al., 1991] makes an attempt to incorporate heating of the plasma sheet population into this class of model. We find that only 60% of all plasma sheet events show densities greater than that in the adjacent lobe, as would be expected on the basis of mass conservation in the reconnection models. In addition, comparison of the distribution of the predicted velocities with the observed velocities shows that for a significant proportion of events, neither reconnection model accurately predicts the observed flows. In fact, the plasma sheet is most often seen to be in a quasi-stagnant state. During periods of tailward convection, the heated model appears the more appropriate for describing plasma sheet flow due to reconnection. However, in the case of the heated (simple) model, at least 50% (60%) of tailward flows are observed at velocities well below the predictions. These events also have positive B z on average and may therefore be associated with the tailward transport of closed flux tubes. They are also associated with geomagnetically quiet times. Such events are seen at all distance between-60 and-240 R• in the plasma sheet. These results strongly support previous suggestions that even at large downtail distances, the observed plasma sheet flows are often driven by processes other than reconnection. While the models correctly predict a significant proportion of the flow speeds for tailward flows, the earthward flow events are not well described by either model. In summary, 40% of the data set cannot be described by the reconnection models due to mass conservation requirements, and at most only 50% of the tailward flow speeds are in agreement with the predictions of these models. As such, only 30% of tailward flow events are well described by the models presented here.