Androdioecy is an uncommon form of reproduction in which males coexist with hermaphrodites. Androdioecy is thought to be difficult to evolve in species that regularly inbreed. The freshwater shrimp Eulimnadia texana has recently been described as both androdioecious and highly selfing and is thus anomalous. Inbreeding depression is one factor that may maintain males in these populations. Here we examine the extent of "late" inbreeding depression (after sexual maturity) in these clam shrimp using two tests: (1) comparing the fitness of shrimp varying in their levels of individual heterozygosity from two natural populations that differ in overall genetic diversity; and (2) specifically outcrossing and selfing shrimp from these same populations and comparing fitness of the resulting offspring. The effects of inbreeding differed within each population. In the more genetically diverse population, fecundity, size, and mortality were significantly reduced in inbred shrimp. In the less genetically diverse population, none of the fitness measures was significantly lowered in selfed shrimp. Combining estimates of early inbreeding depression from a previous study with current estimates of late inbreeding depression suggests that inbreeding depression is substantial (delta = 0.68) in the more diverse population and somewhat lower (delta = 0.50) in the less diverse population. However, given that males have higher mortality rates than hermaphrodites, neither estimate of inbreeding depression is large enough to account for the maintenance of males in either population by inbreeding depression alone. Thus, the stability of androdioecy in this system is likely only if hermaphrodites are unable to self-fertilize many of their own eggs when not mated to a male or if male mating success is generally high (or at least high when males are rare). Patterns of fitness responses in the two populations were consistent with the hypothesis that inbreeding depression is caused by partially recessive deleterious alleles, although a formal test of this hypothesis still needs to be conducted.