The evolution of host specificity remains a central issue in the study of host-parasite relationships. Here we tackle three basic questions about host specificity using data on host use by fleas (Siphonaptera) from 21 geographical regions. First, are the host species exploited by a flea species no more than a random draw from the locally available host species, or do they form a taxonomically distinct subset? Using randomization tests, we showed that in the majority of cases, the taxonomic distinctness (measured as the average taxonomic distances among host species) of the hosts exploited by a flea is no different from that of random subsets of hosts taken from the regional pool. In the several cases where a difference was found, the taxonomic distinctness of the hosts used by a flea was almost always lower than that of the random subsets, suggesting that the parasites use hosts within a narrower taxonomic spectrum than what is available to them. Second, given the variation in host specificity among populations of the same flea species, is host specificity truly a species character? We found that host specificity measures are repeatable among different populations of the same flea species: host specificity varies significantly more among flea species than within flea species. This was true for both measures of host specificity used in the analyses: the number of host species exploited, and the index measuring the average taxonomic distinctness of the host species and its variance. Third, what causes geographical variation in host specificity among populations of the same flea species? In the vast majority of flea species, neither of our two measures of host specificity correlated with either the regional number of potential host species or their taxonomic distinctness, or the distance between the sampled region and the center of the flea's geographical range. However, in most flea species host specificity correlated with measures of the deviation in climatic conditions (precipitation and temperature) between the sampled region and the average conditions computed across the flea's entire range. Overall, these results suggest that host specificity in fleas is to a large extent phylogenetically constrained, while still strongly influenced by local environmental conditions.