Standard matching models of unemployment assume that workers and job ßows are identical. This is in stark contrast to empirical evidence that job ßows in fact only account for a fraction of worker ßows, that unemployment exits only account for a fraction of hires and that these fractions vary over the cycle. In this paper, we develop and calibrate a model based on the Mortensen and Pissarides approach but that emphasises this issue.We show that this matters -that it has very different implications for our view of unemployment dynamics.The key features of our model relate to the search options of the worker, and the job creation decision by Þrms. We allow workers to search whilst employed, and Þrms to re-advertise jobs that have been quit from. This leads us to use a different job creation process, whereby potential vacancies, or job 'ideas', arise at a Þnite rate per period over a range of idiosyncratic productivities. In the standard setting, there is an unlimited supply of potential vacancies at the top idiosyncratic productivity.The main results are as follows. First, the presence of on-the-job search has a substantial impact on labour market equilibrium, whereby equilib- * An earlier version of this paper was called "Unemployment equilibrium and on-the-job search". 1 rium unemployment is lower and exhibits a higher turnover rate. On-thejob search renders the unemployment inßow rate more sensitive to the cycle: in all cases, the inßow rate is found to be more cyclically sensitive than the outßow rate, suggesting that most unemployment dynamics occur through this channel. This conÞrms empirical results for Great Britain (Burgess and Turon (2005)). Second, our model offers some insight into a (two-way) relationship between job-to-job ßows, which drives the difference between worker and job ßows, and the extent of wage dispersion. More wage dispersion increases the incentive to search on-the-job and more on-the-job search widens the range of viable productivities and leads to lower wages at the bottom of the wage distribution, thereby increasing wage dispersion.Third, changes in the model's exogenous parameters impact unemployment to a considerable degree by changing the level of employed job search.