Naturally time-averaged accumulations of skeletal remains—death assemblages—provide reliable, albeit temporally coarse, information on the species composition and structure of communities in diverse settings, and their mismatch with local living communities usually signals recent human-driven ecological change. Here, we present the first test of live–dead mismatch as an indicator of human stress using ostracodes. On three islands along a gradient of human population density in the Bahamas, we compared the similarity of living and death assemblages in 10 lakes with relatively low levels of human stress to live–dead similarity in 11 physically comparable lakes subject to industrial, agricultural, or other human activities currently or in the past. We find that live–dead agreement in pristine lakes is consistently excellent, boding well for using death assemblages in modern-day and paleolimnological biodiversity assessments. In most comparison of physically similar paired lakes, sample-level live–dead mismatch in both taxonomic composition and species' rank abundance is on average significantly greater in the stressed lakes; live–dead agreement is not lower in all samples from stressed lakes, but is more variable. When samples are pooled for lake-level and island-level comparisons, stressed lakes still yield lower live–dead agreement, but the significance of the difference with pristine lakes decreases—species that occur dead-only (or alive-only) in one sample are likely to occur alive (or dead) in other samples. Interisland differences in live–dead agreement are congruent with, but not significantly correlated with, differences in human population density. This situation arises from heterogeneity in the timing and magnitudes of stresses and in the extent of poststress recovery. Live–dead mismatch in ostracode assemblages thus may be a reliable indicator of human impact at the sample level with the potential to be a widely applicable tool for identifying impacted habitats and, perhaps, monitoring the progress of their recovery.