Social integration has been linked to health ever since Emile Durkheim’s study of suicide more than a century ago. Focusing on all-cause mortality as the outcome, we review the epidemiological studies of social networks and health. We update the six previous systematic reviews and meta-analyses on the topic. Social relationships have beneficial impacts on longevity, which are quite robust across different structural domains (such as social networks and social participation) as well as functional domains (such as social support). However, once we ‘drill down’ to a more granular level, some interesting nuances emerge in the overall association between social relationships and health. There is considerable heterogeneity in the magnitude (and occasionally even the direction) of associations according to age-group, gender, source of social relations, as well as contextual characteristics (such as geography and culture). Evidence also suggests potential heterogeneity according to the measure of social relationships used – for example, objective versus subjective social isolation, perceived versus received social support, giving versus receiving support, and positive versus negative social interactions. We discuss the challenges in inferring causality from the observational evidence, and briefly review the potential biological mechanisms underpinning the relation between social integration and mortality.