It is well-known that segregation can already emerge if individuals act upon weak preferences to associate with similar others. Yet, evidence on how such ingroup preferences compare across social settings and identity dimensions remains lacking. To address this gap, we conducted three large-scale, preregistered conjoint experiments in the Netherlands on choices of neighborhoods and civic organizations, which enable us to isolate ingroup preferences from other drivers of segregation. We find that ingroup preferences are strong and widespread, showing up in both settings and across all studied dimensions (age, ethnicity, education). The exception is that lower-educated individuals display no meaningful ingroup preferences along educational lines, whereas higher-educated individuals do. Ingroup preferences are strongest among individuals with little exposure to outgroups in real life and are independent of the expected intensity of contact. Our results demonstrate that preferences for similarity play a critical role in sustaining segregation, over and above opportunity structures.