In September 2009, an outbreak of Salmonella enterica serovar Enteritidis affected 327 of 1419 inmates at a London prison. We applied a cohort design using aggregated data from the kitchen about portions of food distributed, aligned this with individual food histories from 124 cases (18 confirmed, 106 probable) and deduced the exposures of those remaining well. Results showed that prisoners eating egg cress rolls were 26 times more likely to be ill [risk ratio 25·7, 95% confidence interval (CI) 15·5–42·8, P<0·001]. In a case/non-case multivariable analysis the adjusted odds ratio for egg cress rolls was 41·1 (95% CI 10·3–249·7, P<0·001). The epidemiological investigation was strengthened by environmental and microbiological investigations. This paper outlines an approach to investigations in large complex settings where aggregate data for exposures may be available, and led to the development of guidelines for the management of future gastrointestinal outbreaks in prison settings.