There is increasing evidence that our societies are polarizing. Most studies focus on labour market and educational outcomes and show a socioeconomic polarization of the bottom and top ends of the population distribution. Processes of social polarization have a spatial dimension which should be visible in the changing mosaic of neighbourhoods in cities. Many studies treat neighbourhoods as more or less static entities, but urban researchers are now increasingly interested in neighbourhood trajectories, moving away from point-in-time measures and enabling a close examination of processes of change. Sequence analysis allows for a visualization of complete trajectories, and is therefore gaining popularity in the social sciences. However, sequence analysis is mainly a descriptive method and statisticians have argued for the use of a tree-structured discrepancy analysis to examine to what extent outcome variability can be explained by a set of predictors. This paper offers a first empirical application of sequence analysis combined with a tree-structured discrepancy analysis. This paper contributes to the debate on urban renewal programs by offering a unique viewpoint on longitudinal neighbourhood change. Our findings show a clear pattern of socio-spatial polarization in Dutch cities, raising questions about the effects of area-based policies and the importance of path-dependency.