Linking genetic diversity to extinction is a common goal in genomic studies. Recently, a debate has arisen regarding the importance of genetic variation in conservation as some studies have failed to find associations between genome- wide genetic diversity and extinction risk. However, only rarely are genetic diversity and fitness measured together in the wild, and typically demographic history and environment are ignored. It is therefore difficult to infer whether a lack of an association is real or obscured by confounding factors. To address these shortcomings, we analyzed genetic data from 7,501 individuals with extinction data from 279 meadows and mortality of 1,742 larval nests in a butterfly metapopulation. We found a strong negative association between genetic diversity and extinction when considering only heterozygosity in models. However, this association disappeared when accounting for ecological covariates, suggesting a confounding between demography and genetics and a more complex role for heterozygosity in extinction risk. Modeling interactions between heterozygosity and demographic variables revealed that associations between extinction and heterozygosity were context- dependent. For example, extinction declined with increasing heterozygosity in large, but not currently small populations, although negative associations between heterozygosity, extinction, and mortality were detected in small populations with a recent history of decline. We conclude that low genetic diversity is an important predictor of extinction, predicting >25% increase in extinction beyond ecological factors in certain contexts. These results highlight that inferences about the importance of genetic diversity for population viability should not rely on genomic data alone but require investments in obtaining demographic and environmental data from natural populations.