Background. Structural characteristics of social networks such as primary group size have received less attention than measures of perceived social support. Previous research suggests that associations between social network size and later common mental disorder status may differ according to sex and initial mental state.

Method. Adults participating in the 2000 British National Household Survey of psychiatric morbidity were randomly selected for follow-up 18 months later. The revised Clinical Interview Schedule (CIS-R) and the Interview Measure of Social Relations (IMSR) were administered at baseline and follow-up. Primary group size was defined as the total number of close relatives and friends. A four-level scale of common mental disorder was modelled with ordinal logistic regression, based on weighted data (n=2413).

Findings. After adjusting for confounders, a primary group size of three or less at time 1 predicted worse mental health at time 2. This effect was greatest in men who were initially non-cases at baseline (averaged odds 4·5) and in women who were initially cases at baseline (average odds 2·9). Primary group size at time 2 was significantly predicted by level of common mental disorder at time 1 in women but not in men. Thus, confounding by baseline disorder does not explain risk of developing poor mental health in socially isolated men.

Conclusion. This study replicates the strong effects of primary group size on future mental health that emerge when men and women are studied separately and when subjects are categorized according to baseline mental health status.