The ability to survive in a terrestrial environment was a major evolutionary hurdle for animals that, once passed, allowed the diversification of most arthropod and vertebrate lineages. Return to a truly aquatic lifestyle has occurred only rarely among terrestrial lineages, and is generally associated with modifications of the respiratory system to conserve oxygen and allow extended periods of apnea. Among chelicerates, in particular spiders, where the circulatory system also serves as a hydrostatic skeleton, very few taxa have exploited aquatic environments, though these environments are abundant and range from freshwater ponds to the marine intertidal and relictual (salt) lakes. The traditional systematic positions of the taxa inhabiting these environments are controversial. Partitioned Bayesian analysis using a doublet model for stems in the nearly complete 18S rRNA gene (~1800 nt) and in the D2 and D3 regions of the 28S rRNA gene (~690 nt), and standard models for loops and full protein-coding histone H3 (349 nt) partitions (totalling 3133 bp when aligned) of dictynoid spiders and related lineages revealed that the only truly aquatic spider species, Argyroneta aquatica (Clerck, 1767) (Cybaeidae Banks, 1892), belongs in a clade containing other taxa with unusual habitat affinities related to an aquatic existence, including occupation of semi-aquatic (intertidal) areas (Desidae Pocock, 1985: Paratheuma spp.) and highly alkaline salt-crusts (Dictynidae O. Pickard-Cambridge, 1871: Saltonia incerta (Banks, 1898)). In a contrasting pattern, other spiders that also occupy intertidal zones, including some other members of the family Desidae (Desis spp., Badumna longinqua (L. Koch, 1867)), are an independently derived clade found primarily in the southern hemisphere. Use of the doublet model reduced some branch-support values in the single-gene trees for rRNA data, but resulted in a robust combined-data phylogeny from 18S rRNA, 28S rRNA, and histone H3. This combination of results – reduction in support in single-gene trees and gain in support in combined-data trees –is consistent with use of the doublet model reducing problematic signal from non-independent base pairs in individual data partitions, resulting in improved resolution in the combined-data analyses.