The mismatch discrimination potential of probes in fluorescence in situ hybridization can be defined as the difference between the melting formamide points of perfect complementary and mismatched duplexes (Delta[FA](m)). Using a combined experimental and theoretical approach, Delta[FA](m) was determined for a set of 35 mismatched probes targeting seven locations in the 16S rRNA of Escherichia coli. The mismatches were created by changing single nucleotides on the probes, while maintaining the target unmodified. Estimated Delta[FA](m) values were used to systematically evaluate four predictors of mismatch stability: weighted mismatch (WM) scores from the software arb, published statistical summary of microarray hybridizations, free energy of mismatch stability (DeltaDeltaG degrees (1)) and theoretical Delta[FA](m) estimations obtained with a thermodynamic model. Based on the predictors' ability to explain variability in Delta[FA](m) and to discriminate weak mismatches from strong ones, DeltaDeltaG degrees (1) and WM scores from arb (with an updated set of relative strength parameters) were demonstrated to be adequate estimators of mismatch stability, with DeltaDeltaG degrees (1) offering the benefit of capturing the variability associated with nearest-neighbour effects and being compatible with thermodynamic models of in situ hybridization. The use of DeltaDeltaG degrees (1) and WM in probe design was illustrated as a tool that complements experimental design approaches.