The dilution effect, that high host species diversity can reduce disease risk, has attracted much attention in the context of global biodiversity decline and increasing disease emergence. Recent studies have criticized the generality of the dilution effect and argued that it only occurs under certain circumstances. Nevertheless, evidence for the existence of a dilution effect was reported in about 80% of the studies that addressed the diversity–disease relationship, and a recent meta-analysis found that the dilution effect is widespread. We here review supporting and critical studies, point out the causes underlying the current disputes. The dilution is expected to be strong when the competent host species tend to remain when species diversity declines, characterized as a negative relationship between species’ reservoir competence and local extinction risk. We here conclude that most studies support a negative competence–extinction relationship. We then synthesize the current knowledge on how the diversity–disease relationship can be modified by particular species in community, by the scales of analyses, and by the disease risk measures. We also highlight the complex role of habitat fragmentation in the diversity–disease relationship from epidemiological, evolutionary and ecological perspectives, and construct a synthetic framework integrating these three perspectives. We suggest that future studies should test the diversity–disease relationship across different scales and consider the multiple effects of landscape fragmentation.