The decay of HIV-1-infected cell populations after treatment with antiretroviral therapy has been measured using simple exponential decay models. These models are unlikely to be realistic over periods longer than a few months, however, because the population dynamics of HIV are complex. We considered an alternate model developed by Perelson and colleagues that extends the standard model for biphasic viral load decline and allows for nonlinear log decay of infected cell populations. Using data from 6 children on highly active antiretroviral therapy (HAART) and a single parameter in the new model, the assumption of log linear decay of infected cell populations is tested. Our analysis indicates that the short-lived and long-lived infected cell populations do not decay according to a simple exponential model. Furthermore, the resulting estimates of time to eradication of infected cell compartments are dramatically longer than those previously reported (eg, decades vs. years for long-lived infected cell populations and years vs. weeks for short-lived infected cell populations). Furthermore, estimates of the second-phase decay rates are significantly different than 0 for most children when obtained using the Perelson biphasic decay model. In contrast, this rate is not significantly different than 0 when the density-dependent decay model is used for parameter estimation and inference. Thus, the density-dependent decay model but not the simple exponential decay model is consistent with recent data showing that even under consistent HAART-mediated suppression of viral replication, decay rates of infected cell reservoirs decay little over several years. This suggests that conclusions about long-term viral dynamics of HIV infection based on simple exponential decay models should be carefully re-evaluated.