Many methods for typing microbial strains are not 100% reproducible. This can create problems when deciding whether different groups of isolates are really distinct or represent typing errors or variation of a single strain. Neither hierarchical clustering nor iterative partitioning methods are suited for analysing such data. A novel iterative partitioning method is described which allows for the uncertainty of the typing method in use. Before grouping strains, the maximum dimension of the groups is set based on a previous knowledge of the typing method's reproducibility. Isolates are only allocated to a group if they differ from that group's typical strain type by less than the number of reaction differences required to distinguish between two strains. In a series of Monte Carlo studies the accuracy of strain allocation was found to be very good, even when the two groups were situated close to each other.