Although interest in monopsonistic influences on labour market outcomes has revived in recent years, only a few empirical studies provide direct evidence on it. This paper analyses empirically the effect of monopsony power on pay structure, using a direct measure of labour market 'thinness'. We find that having fewer competitors for skilled labour is associated at the level of the establishment with lower pay for both skilled labour and trainees, but not for unskilled labour. These findings have potentially important implications for the economic theory of training, as most recent models assume that skilled pay is set monopsonistically but both unskilled and trainee pay are determined competitively. Our results support those assumptions for skilled pay and unskilled pay, but not for trainee pay.