The U.S. labor market has been experiencing unprecedented high average unemployment duration. The shift in the unemployment duration distribution can be traced back to the early nineties. In this study, censored quantile regression methods are employed to analyze the changes in the US unemployment duration distribution. We explore the decomposition method proposed by Machado and Mata (2005) to disentangle the contribution of the changes generated by the covariate distribution and by the conditional distribution. The data used in this inquiry are taken from the nationally representative Displaced Worker Surveys of 1988 and 2008. We provide evidence that the change in the unemployment duration distribution is mainly produced by a compositional change of the covariates. Apart from the effect of economic improvement we find that the sensitivity of joblessness duration to education and the unemployment rate were the main forces behind the increase of the unemployment duration, in the last twenty years.