The unusual activity of supported Au catalysts for low-temperature CO oxidation has generated world-wide effort to understand its origin. Although the data point to the importance of Au particle size, there is still controversy as to whether the activity is due to size-electronic effect, the presence of low-coordination surface atoms or atoms at the perimeter, or whether it is due to the presence of Au ions. We have attempted to address these questions by following the development of catalytic activity starting from the catalyst precursors and the effect of halide poisoning. Starting from Au oxyhydroxy complexes deposited on a support, catalytic activity is developed in conjunction with reduction of the complex to metallic Au particles. On the contrary, much reduced activity observed on a halide-containing sample is accompanied by the presence of unreduced Au complexes. On a fully reduced, highly active sample, the activity can be effectively suppressed when Br ions are bonded to Au atoms without oxidizing Au. The Br ions preferentially poison the active sites instead of CO adsorption sites. The data supports the model that either perimeter sites or low coordination sites accounts for the high activity of Au catalysts for CO oxidation.