Despite extensive paleoecological analyses of spatial and temporal turnover in species composition, the fidelity with which time-averaged death assemblages capture variation in species composition and diversity partitioning of living communities remains unexplored. Do death assemblages vary in composition between sites to a lesser degree than do living assemblages, as would be predicted from time-averaging? And is the higher number of species observed in death relative to living assemblages reduced with increasing spatial scale? We quantify the preservation of spatial and temporal variation in species composition using 11 regional data sets based on samples of living molluscan communities and their co-occurring time-averaged death assemblages. (1) Compositional dissimilarities among living assemblages (LA) within data sets are significantly positively rank-correlated to dissimilarities among counterpart pairs of death assemblages (DA), demonstrating that pairwise dissimilarity within a study area has a good preservation potential in the fossil record. Dissimilarity indices that downplay the abundance of dominant species return the highest live-dead agreement of variation in species composition. (2) The average variation in species composition (average dissimilarity) is consistently smaller in DAs than in LAs (9 of 11 data sets). This damping of variation might arise from DAs generally having a larger sample size, but the reduction by ∼10–20% mostly persists even in size-standardized analyses (4 to 7 of 11 data sets, depending on metric). Beta diversity expressed by the number of compositionally distinct communities is also significantly reduced in death assemblages in size-standardized analyses (by ∼25%). This damping of variation and reduction in beta diversity is in accord with the loss of temporal resolution expected from time-averaging, without invoking taphonomic bias (from differential preservation or postmortem transportation) or sample-size effects. The loss of temporal resolution should directly reduce temporal variation, and assuming time-for-space substitution owing to random walk within one habitat and/or temporal habitat shifting, it also decreases spatial variation in species composition. (3) DAs are more diverse than LAs at the alpha scale, but the difference is reduced at gamma scales because partitioning of alpha and beta components differs significantly between LAs and DAs. This indicates that the effects of time-averaging are reduced with increasing spatial scale. Thus, overall, time-averaged molluscan DAs do capture variation among samples of the living assemblage, but they tend to damp the magnitude of variation, making them a conservative means of inferring change over time or variation among regions in species composition and diversity. Rates of temporal and spatial species turnover documented in the fossil record are thus expected to be depressed relative to the turnover rates that are predicted by models of community dynamics, which assume higher temporal resolution. Finally, the capture by DAs of underlying variation in the LA implies little variation in the net preservation potential of death assemblages across environments, despite the different taphonomic pathways suggested by taphofacies studies.