OBJECTIVE: To compare the health outcomes in sheltered and evacuated populations after a chemical incident in a plastics factory. DESIGN: Cross sectional survey. SETTING: Urban area in southwest England. PARTICIPANTS: 1750 residents from the area exposed to the chemical smoke, of which 472 were evacuated and the remaining 1278 were advised to shelter indoors. MAIN OUTCOME MEASURE: Number of adverse health symptoms. A case was defined by the presence of four or more symptoms. MAIN RESULTS: 1096 residents (63%; 299 evacuated, 797 sheltered) provided data for analyses. The mean symptom score and proportion of cases were higher in evacuated people than in the sheltered population (evacuated: symptom score 1.9, cases 19.7% (n = 59); sheltered: symptom score 1.0, cases 9.5% (n = 76); P < 0.001 for both). The difference between the two groups attenuated markedly at the end of two weeks from the start of the incident. The two main modifiable risk factors for the odds of becoming a case were evacuation (odds ratio 2.5, 95% confidence interval 1.7 to 3.8) and direct exposure to smoke for more than two hours on the first day of the incident (2.0, 1.7 to 2.3). The distance of residence from the factory or level of exposure before intervention (first six hours) had little effect on the odds of a person becoming a case. CONCLUSIONS: Sheltering may have been a better protective action than evacuation in this chemical incident, which is consistent with the prevailing expert view. Although this study has limitations, it is based on a real event. Evacuations carry their own risks and resource implications; increased awareness may help to reduce unnecessary evacuations in the future.