Surveillance of antibiotic resistance can be undertaken by compilation of routine data or by central testing of isolates. Routine results can be obtained cheaply and in sufficient quantities for correlation with population and prescribing denominators but there is concern about their quality. As one of a series of ongoing studies to assess this quality, we compared the proportions of resistance amongst Escherichia coli from patients with bacteraemia or meningitis between 1991 and 1997 (i) as recorded in routine data reported to the PHLS and (ii) as found in tests performed at the PHLS Laboratory of Enteric Pathogens (LEP). These two data sets both showed an overall upward trend in the proportion of isolates resistant to ampicillin, trimethoprim, gentamicin and ciprofloxacin. The average annual percentage increase in resistance was estimated in separate logistic regression models, and 95% confidence intervals (CI) were determined. The annual percentage increases in the proportions of isolates reported resistant were similar in the two data sets for trimethoprim, gentamicin and ciprofloxacin but differed for ampicillin. The upward trends were statistically significant except for gentamicin resistance in the LEP data set, where the 95% CI straddled zero. The proportions of resistant isolates for each antibiotic in the two data sets each year were in poorer agreement than the trends; however, the 95% CI of the difference of proportions resistant between the routine and LEP data sets straddled zero in 4 or 5 of the 7 years studied. Some discrepancies might be explained by geographical bias in the sampling or by differences in definitions of resistance. Thus (i) the proportion of resistant isolates tested at LEP almost always fell within the ranges bounded by the highest and lowest proportions for individual Regional Health Authorities, as recorded in the routine data, and (ii) the fact that LEP consistently recorded less gentamicin resistance but more ciprofloxacin resistance than the routine could be explained by breakpoint differences. We conclude that routine susceptibility data for ampicillin, ciprofloxacin, gentamicin and trimethoprim appear sound for E. coli and might be suitable for correlation with other data, e.g. for prescribing.