Extended-spectrum beta-lactamases (ESBLs) are enzymes found in gram-negative bacilli that mediate resistance to extended-spectrum cephalosporins and aztreonam. In 1999, the National Committee for Clinical Laboratory Standards (NCCLS) published methods for screening and confirming the presence of ESBLs in Klebsiella pneumoniae, Klebsiella oxytoca, and Escherichia coli. To evaluate the confirmation protocol, we tested 139 isolates of K. pneumoniae that were sent to Project ICARE (Intensive Care Antimicrobial Resistance Epidemiology) from 19 hospitals in 11 U.S. states. Each isolate met the NCCLS screening criteria for potential ESBL producers (ceftazidime [CAZ] or cefotaxime [CTX] MICs were > or =2 microg/ml for all isolates). Initially, 117 (84%) isolates demonstrated a clavulanic acid (CA) effect by disk diffusion (i.e., an increase in CAZ or CTX zone diameters of > or =5 mm in the presence of CA), and 114 (82%) demonstrated a CA effect by broth microdilution (reduction of CAZ or CTX MICs by > or =3 dilutions). For five isolates, a CA effect could not be determined initially by broth microdilution because of off-scale CAZ results. However, a CA effect was observed in two of these isolates by testing cefepime and cefepime plus CA. The cefoxitin MICs for 23 isolates that failed to show a CA effect by broth microdilution were > or =32 microg/ml, suggesting either the presence of an AmpC-type beta-lactamase or porin changes that could mask a CA effect. By isoelectric focusing (IEF), 7 of the 23 isolates contained a beta-lactamase with a pI of > or =8.3 suggestive of an AmpC-type beta-lactamase; 6 of the 7 isolates were shown by PCR to contain both ampC-type and bla(OXA) genes. The IEF profiles of the remaining 16 isolates showed a variety of beta-lactamase bands, all of which had pIs of < or =7.5. All 16 isolates were negative by PCR with multiple primer sets for ampC-type, bla(OXA), and bla(CTX-M) genes. In summary, 83.5% of the K. pneumoniae isolates that were identified initially as presumptive ESBL producers were positive for a CA effect, while 5.0% contained beta-lactamases that likely masked the CA effect. The remaining 11.5% of the isolates studied contained beta-lactamases that did not demonstrate a CA effect. An algorithm based on phenotypic analyses is suggested for evaluation of such isolates.