The site stripping for clock detection procedure was implemented in the recently developed maximum likelihood framework for estimating evolutionary rates and divergence times in measurably evolving populations. The method was used to investigate the effect of rate variability on estimating divergence times in non-clock-like trees for human immunodeficiency viruses and hepatitis C viruses. We validate our approach by comparing dated coalescent nodes in molecular phylogenies with known dates of transmission. Our method was able to rapidly recover clock-like behavior and to indicate the presence and direction of a bias when estimates of divergence times using the unstripped data were flawed.