In this paper we develop a model of workers' quit decisions. Using panel data on a sample of US Federal government employees we use maximum-likelihood techniques to determine how much of the observed decline in quits with job tenure is a result of declining individual quit propensities (`state dependence') rather than dynamic sample self-selection (`heterogeneity'). The latter is found to be much more important than the former. The effects of relative Federal government wages and other variables on quits are also estimated. The results are useful for analysing Federal compensation policies.