Colicine production and sensitivity were determined for strains of E. coli of defined serotype isolated from patients in a medical ward. The strains could be assigned to 19 colicine types as determined by production patterns. No correlation was observed between colicine production or sensitivity and serotype. Colicine typing proved to be stable and hence reliable as an epidemiological marker. However, only one-third of the strains of E. coli were typable. We used this method in an epidemiological investigation and found some evidence that E. coli of animal origin were reaching hospital patients via food.