The structure of methylaluminoxane (MAO), used as a cocatalyst for olefin polymerization, has been investigated by Raman and in situ IR spectroscopy, polymerization experiments, and density functional calculations. From experimental results, a number of quantum chemical calculations, and bonding properties of related compounds, we have suggested a few Me18Al12O9 cage structures, including a highly regular one with C-3h symmetry, which may serve as models for methylaluminoxane solutions. The cages themselves are rigid but may contain up to three bridging methyl groups on the cage surfaces that are labile and reactive. Bridging methyls were substituted with Cl atoms to form a compound otherwise similar to MAO. Chlorinated MAO is unable to activate a metallocene catalyst, even in the presence of trimethylaluminum (TMA), but allows subsequent activation by regular MAO. With bis(pentamethylcyclopentadienyl)zirconium dichloride, MAO and TMA seem to influence chain termination independently. Several findings previously poorly explained are rationalized with the new model, including the observed lack of reaction products with excess TMA.