Using a semi-structural approach based on a dynamic monopsony model, we examine to what extent workers performing different job tasks are exposed to different degrees of monopsony power, and whether these differences in monopsony power have changed over the last 30 years. We find that workers performing mostly non-routine cognitive tasks are exposed to a higher degree of monopsony power than workers performing routine or non-routine manual tasks. Job-specific human capital and non-pecuniary job characteristics are the most likely explanations for this result. We find no evidence that labour market polarisation has increased monopsony power over time.