With the objective of analyzing which kind of reference data is appropriate for benchmarking quantum chemical approaches for transition metal compounds, we present the following, (a) a collection of 60 transition metal diatomic molecules for which experimentally derived dissociation energies, equilibrium distances, and harmonic vibrational frequencies are known and (b) a composite computational approach based on coupled-cluster theory with basis set extrapolation, inclusion of core-valence correlation, and corrections for relativistic and multireference effects. The latter correction was obtained from internally contracted multireference coupled-cluster (icMRCC) theory. This composite approach has been used to obtain the dissociation energies and spectroscopic constants for the 60 molecules in our data set. In accordance with previous studies on a subset of molecules, we find that multireference corrections are rather small in many cases and CCSD(T) can provide accurate reference values, if the complete basis set limit is explored. In addition, the multireference correction improves the results in cases where CCSD(T) is not a good approximation. For a few cases, however, strong deviations from experiment persist, which cannot be explained by the remaining error in the computational approach. We suggest that these experimentally derived values require careful revision. This also shows that reliable reference values for benchmarking approximate computational methods are not always easily accessible via experiment and accurate computations may provide an alternative way to access them. In order to assess how the choice of reference data affects benchmark studies, we tested 10 DFT functionals for the molecules in the present data set against experimental and calculated reference values. Despite the differences between these two sets of reference values, we found that the ranking of the relative performance of the DFT functionals is nearly independent of the chosen reference.