Using experimentally induced disruptive selection, we tested two hypotheses regarding the evolution of specialization in parasites. The ‘trade-off’ hypothesis suggests that adaptation to a specific host may come at the expense of a reduced performance when exploiting another host. The alternative ‘relaxed selection’ hypothesis suggests that the ability to exploit a given host would deteriorate when becoming obsolete. Three replicate populations of a flea Xenopsylla ramesis were maintained on each of two rodent hosts, Meriones crassus and Dipodillus dasyurus, for nine generations. Fleas maintained on a specific host species for a few generations substantially decreased their reproductive performance when transferred to an alternative host species, whereas they generally did not increase their performance on their maintenance host. The results support the ‘relaxed selection’ hypothesis of the evolution of ecological specialization in haematophagous ectoparasites, while suggesting that trade-offs are unlikely drivers of specialization. Further work is needed to study the extent by which the observed specializations are based on epigenetic or genetic modifications.