Couples often resemble each other in characteristics like depression, but the reasons for this homogamy (i.e., similarity) remain unclear. We investigated two potential mechanisms: preference for a self-similar partner and convergence (i.e., increasing similarity) over time. In a nationally representative sample of 2,793 Czech individuals who we surveyed three times in one year, we examined self-reports of participants’, their ideal partners’, and their actual partners’ “pessimism and depressiveness”. Participants preferred partners less depressive than themselves, yet their actual partners were more depressive than desired. Those who ended their relationships showed a greater ideal-versus-actual partner discrepancy than those who stayed together. In stable relationships, individuals adjusted their ideal preferences to align more closely with their actual partners over time. We identified four relationship classes with latent class growth modeling based on self and partner evaluations: both non-depressive, both depressive, self depressive and partner non-depressive, and self non-depressive and partner depressive. Romantic relationships were most stable when both partners were non-depressive and most likely to dissolve when both were depressive. While we failed to detect convergence overall, we found it within heterogamous (i.e., dissimilar) classes. Overall, our findings suggest that homogamy and heterogamy in depressiveness are complexly associated with relationship maintenance.