We revisited the role of dissimilarity of host assemblages in shaping dissimilarity of flea assemblages using a non-linear approach. Generalized dissimilarity models (GDMs) were applied using data from regional surveys of fleas parasitic on small mammals in four biogeographical realms. We compared (1) model fit, (2) the relative effects of host compositional and phylogenetic turnover and geographic distance on flea compositional and phylogenetic turnover, and (3) the rate of flea turnover along gradients of host turnover and geographic distance with those from earlier application of a linear approach. GDMs outperformed linear models in explaining variation in flea species turnover and host dissimilarity was the best predictor of flea dissimilarity, irrespective of scale. The shape of the relationships between flea compositional turnovers along host compositional turnover was similar in all realms, whereas turnover along geographic distance differed among realms. In contrast, the rate of flea phylogenetic turnover along gradients of host phylogenetic turnover differed among realms, whereas flea phylogenetic turnover did not depend on geographic distance in any realm. We demonstrated that a non-linear approach (a) explained spatial variation in parasite community composition better than and (b) revealed patterns that were obscured by earlier linear analyses.