This study evaluates whether estimated multidrug resistance (MDR) levels are dependent on the design of the surveillance system when using routine microbiological data. We used antimicrobial resistance data from the Antibiotic Resistance and Prescribing in European Children (ARPEC) project. The MDR status of bloodstream isolates of Escherichia coli, Klebsiella pneumoniae and Pseudomonas aeruginosa was defined using European Centre for Disease Prevention and Control (ECDC)-endorsed standardised algorithms (non-susceptible to at least one agent in three or more antibiotic classes). Assessment of MDR status was based on specified combinations of antibiotic classes reportable as part of routine surveillance activities. The agreement between MDR status and resistance to specific pathogen-antibiotic class combinations (PACCs) was assessed. Based on all available antibiotic susceptibility testing, the proportion of MDR isolates was 31% for E. coli, 30% for K. pneumoniae and 28% for P. aeruginosa isolates. These proportions fell to 9, 14 and 25%, respectively, when based only on classes collected by current ECDC surveillance methods. Resistance percentages for specific PACCs were lower compared with MDR percentages, except for P. aeruginosa. Accordingly, MDR detection based on these had low sensitivity for E. coli (2-41%) and K. pneumoniae (21-85%). Estimates of MDR percentages for Gram-negative bacteria are strongly influenced by the antibiotic classes reported. When a complete set of results requested by the algorithm is not available, inclusion of classes frequently tested as part of routine clinical care greatly improves the detection of MDR. Resistance to individual PACCs should not be considered reflective of MDR percentages in Enterobacteriaceae.