We report detailed measurements of the thermal and optical properties of the well-known gold-acceptor level in silicon. From a comparison of these properties measured in two types of silicon ${p}^{+}n$ diodes (one fabricated from epitaxially grown silicon and the other from silicon grown by the Czochrolski technique), we find that there are at least two (and perhaps more) different types of gold-acceptor centers in silicon. Measurements of the ratio of the deep-level transient spectroscopy signal due to the gold-donor level to that of the gold acceptor in the same sample show that these two levels are not related to the same gold center, as had previously been believed. A comparison with data in the literature for five other approximately midgap levels in silicon (Ag, Co, Rh, S, and process-induced levels) shows that the electron thermal-emission rates of these are all identical to that of the gold acceptor within experimental error. This suggests that these deep levels have the same underlying defect structure. Comparisons of our measured electron-capture cross sections with those reported in the literature for the gold acceptor show a heretofore unreported correlation between the magnitude of this cross section and the ratio of gold concentration to that of the shallow donor impurity. This suggests that ion pairing between gold acceptors and shallow donors plays a role in determining the capture cross section of the defect. We also have measured for the first time the capture cross sections associated with the gold-donor defect in $n$-type material and find results significantly different from those reported in the literature for $p$-type silicon. The optical cross sections of the gold acceptor show nearly a factor of 10 difference in magnitude and subtle differences in shape between epitaxial and Czochrolski samples. It is therefore possible that oxygen is also playing a role in the gold-based defect complexes. Attempts to obtain the temperature dependence of the gold-acceptor level from thermal capture and emission data proved inconclusive and led to various paradoxes which we discuss. Finally, we discuss possible models for gold-related defect complexes which are consistent with our results and might form the basis for future work.