A subset of Escherichia coli, Klebsiella oxytoca, Klebsiella pneumoniae and Proteus mirabilis isolates collected for the Study for Monitoring Antimicrobial Resistance Trends that were positive for the Clinical and Laboratory Standards Institute (CLSI) extended-spectrum β-lactamase (ESBL) phenotypic confirmatory test (n = 3245) or had an ertapenem MIC of ≥0.5 µg ml(-1) (n = 293), or both (n = 467), were analysed for ESBL genes. Most ESBL phenotype E. coli or K. pneumoniae possessed an ESBL gene (95.8 and 88.4 %, respectively), and this was 93.1 % if carbapenem-non-susceptible K. pneumoniae were removed. This rate was lower for P. mirabilis (73.4 %) and K. oxytoca (62.5 %). Virtually all ESBL-positive isolates (99.5 %) were cefotaxime non-susceptible [CLSI or European Committee on Antimicrobial Susceptibility Testing (EUCAST) breakpoints)]. Fewer isolates (82 %) were ceftazidime non-susceptible (CLSI breakpoints). In addition, 21.1 % of E. coli, 25 % of K. oxytoca and 78.7 % of P. mirabilis isolates were ceftazidime susceptible but ESBL positive. This suggests that CLSI breakpoints for ceftazidime are too high to detect ESBLs. The lower EUCAST breakpoints detected ESBLs in E. coli and K. oxytoca better, but 59.6 % of ESBL-positive isolates of P. mirabilis were ceftazidime susceptible. For isolates with ertapenem MICs ≥0.5 µg ml(-1), more accurate ESBL phenotype analysis was observed for E. coli and K. pneumoniae (sensitivity >95 % for both, specificity 94.4 and 54.1 %, respectively). If carbapenemase-positive K. pneumoniae were excluded, the specificity increased to 78 %. The positive predictive values for the ESBL phenotypic test with E. coli and K. pneumoniae were 97.6 and 81.8 %, respectively, and negative predictive values were 75.9 and 95.2 %, respectively. We therefore suggest that it would be prudent to confirm phenotypic ESBL-positive P. mirabilis, K. pneumoniae and K. oxytoca with molecular analysis.