The polycentric models of the new urban economics (NUE) predict that population density decreases while increasing the distance to employment centres. In contrast with this, some studies have calculated non-significant gradients or even positive ones, which appear to threaten seriously the usefulness of these theoretical models. Does this mean that this theoretical framework should be given up in order to understand the decision-making processes of the actors in a polycentric city and their cumulative effects on urban structure? Or, rather, is it a matter of overcoming problems with the appropriate estimating techniques? This study has tested the effect of decentralised and integrated sub-centres in the Barcelona metropolitan region on population density in 1991 and 2001. From the preliminary results, it is clear that population density increases with distance in a considerable number of the sub-centres that have sprung up as employment has decentralised. It has been detected that this result is due not so much to the higher value of more distant residential land compared with that nearer the employment sub-centres, but to deficiencies in the econometric model used. The problem is that the sub-centres belonging to this group are very close together. Once this is resolved, it is demonstrated that, although distance has less effect on decentralised sub-centres than integrated ones, in both cases the effect is negative; that is, when distance increases, population density is reduced. Therefore, the results obtained are not clearly contrary to the predictions of the theoretical models. © 2008 Urban Studies Journal Limited.