Varying rates of false-positive results of phenotypic extended-spectrum β-lactamase (ESBL) tests have been reported for different methods in different settings, species and geographic locations. This report describes discrepancies in Escherichia coli genotypic and phenotypic ESBL rates observed in a surveillance study of 29 US hospitals that participated in the Study for Monitoring Antimicrobial Resistance Trends (SMART). The ESBL phenotype was determined with the Clinical and Laboratory Standards Institute confirmatory broth microdilution test using cefotaxime and ceftazidime with and without clavulanate. Genes encoding ESBLs, carbapenemases and plasmidic AmpC β-lactamases were detected using a combination of microarray and multiplex PCR assays. Among 168 molecularly characterized phenotypically ESBL-positive E. coli isolates from intra-abdominal infections, 4.8 % were genotypically negative from 2009 to 2012 and 29.5 % in 2013. Because of the high rate of false-positive phenotypic ESBL results in 2013, the 5-year phenotypic ESBL trend was skewed and showed a statistically significant increase (P<0.05) in ESBL-positive E. coli in the USA, which was not seen using the genotypic ESBL rates. The majority of false-positive phenotypic profiles had ceftazidime MICs of 2 µg ml-1 and a ≥3 doubling dilution decrease in MIC for only one of the two antimicrobial agents. False-positive ESBL results can adversely impact epidemiological surveillance and patient care (including inappropriate treatment, unnecessary patient isolation and higher costs). Careful evaluation and comparison of phenotypic and genotypic test results can yield the greatest insight, but the most accurate (and faster) detection of ESBL producers is usually based on molecular data.