PFGE is currently the North American standard for surveillance for Clostridium difficile but lacks discriminatory power to aid outbreak investigation. A further limitation to PFGE is the high baseline rate of the epidemic North American pulsotype (NAP) 1 strain in hospitals. Multilocus variable-number tandem repeat analysis (MLVA) appears to have superior discriminatory power but criteria to define clonality have not been set. We conducted surveillance for toxin-positive C. difficile infection (CDI) at a single academic health sciences centre between September 2009 and April 2010. Seventy-four patient specimens resulting in 86 discrete CDI episodes were subjected to PFGE and MLVA. Results were analysed using Bionumerics software to generate phylogenetic trees and coupled to patient demographic data. Amongst the NAP1 strains, two distinct clusters were identified by MLVA using 90 % similarity as a cut-off by Manhattan distance-based clustering, four clusters using 95 % and seven clusters using 97 %. Population analysis conducted on multiple colonies (n = 25) demonstrated that 1-3 % difference in MLVA types was typical for a single individual. Typing was also conducted in the context of institutional outbreaks (n = 42, three outbreaks) in order to determine clusters within the NAP1 strain. By combining longitudinal surveillance with epidemiological information, single specimen population analysis and typing in the context of institutional outbreaks, we conclude that the use of the Manhattan distance-based clustering with a cut-off of 95-97 % is capable of distinguishing outbreak clones from sporadic isolates.