Most research in automatic music genre recognition
has used the dataset assembled by Tzanetakis et al. \cite{Tzanetakis2002,Tzanetakis2002b}.
The integrity of this dataset, however, has never been analyzed.
We catalog numerous serious problems in this dataset,
including replications, mislabelings, versions, and data corruption.
These problems affect the validity of all results derived from it;
but they also present new challenges,
especially now that researchers are using datasets so large
that manual validation of their integrity is impossible.