Though a lot of work has been done on the distribution of job tenures, we are still uncertain about its main determinants. In this paper, we stress random shocks to match productivity after the start of an employment relation. The specificity of investment makes hiring and separation decisions irreversible.These decisions therefore have an option value. Assumptions on riskneutrality, efficient bargaining, and the efficient resolution of hold up problems allow investment and separation decisions to be analyzed separately from wage setting. The tenure profiles in wages implied by the model fit the observed pattern quite well. The model yields a hump shaped pattern in separation rates, similar to learning models, but with a slowerdecline after the peak. Estimation results using job tenure data from the NLSY support this humped shaped pattern and favor this model above the learning model. We develop a methodology to analyze the decomposition of shocks to match productivity into idiosyncratic and macro-level shocks.When assuming a Last-In-First-Out (LIFO) separation rule, this model of individualemployment relations is embedded in a model of firm level employment, that satisfies Gibrat's law. The LIFO rule is interpreted as an institution protecting the property rights on specific investments of incumbentworkers against hiring new workers by the firm.