In this paper we document the patterns of employment adjustment at the micro-level. We find clear evidence of lumpy adjustment consistent with the presence of non-convexities in the adjustment technology - inaction is pervasive, action spells are short-lived, extreme adjustment episodes occur and are responsible for a non-trivial share of employment adjustment. We also find that the probability of employment adjustment increases with the duration of inaction (positive duration dependence). The skill structure of the workforce, the type of employment contract and the proportion of low tenure workers, which we interpret as proxies for the magnitude of adjustment costs, all influence the probability of adjustment.