SummarySome of the issues in the analysis of durations of unemployment spells are considered, including, in particular, notions of duration dependence and the effect of the sampling scheme on the correct incorporation of incomplete or censored spells. Attention is restricted to analysis of spells overlapping a particular time point, and consideration is given to the situations where one has data only on the backward recurrence times, and where additional follow‐up information on the forward recurrence times is also available. In the former case a quasi‐likelihood approach is suggested as a possible alternative to full parametric modelling under some special conditions, while in the latter case arguments are forwarded in favour of analyses based on the conditional distribution of the forward recurrence times given the backward recurrence times. The methods are briefly illustrated using samples extracted from the Australian Longitudinal Sur