The effects of social network structure, support and physical health status on psychiatric morbidity were investigated among 1415 people over retirement age who took part in three independent but comparable surveys in London (urban area) and Essex (semi rural area). Multivariate analysis showed that the model explained between 14.3% and 28.6% of the variation in psychiatric morbidity in the three samples. Poor health status was a more powerful predictor of psychiatric morbidity than the social network variables. Age and sex contributed little to the model. The model was strongest among the two samples of Hackney respondents.