The mechanism of ethylene polymerization by the widely used Phillips catalyst remains controversial. In this work, we compare initiation, propagation, and termination pathways computationally using small chromasiloxane cluster models for several previously proposed and new mechanisms. Where possible, we consider complete catalytic cycles and compare predicted kinetics, active site abundances, and polymer molecular weights to known properties of the Phillips catalyst. Prohibitively high activation barriers for propagation rule out previously proposed chromacycle ring expansion and Green–Rooney (alternating alkylidene/chromacycle) mechanisms. A new oxachromacycle ring expansion mechanism has a plausible propagation barrier, but initiation is prohibitively slow. On sites with adjacent bridging hydroxyls, either ≡Si(OH)CrII-alkyl or ≡Si(OH)CrIII-alkyl, initiated by proton transfer from ethylene, chain growth by a Cossee–Arlman-type mechanism is fast. However, the initiation step is uphill and extremely slow, so essentially all sites remain trapped in a dormant state. In addition, these sites make only oligomers because when all pathways are considered, termination is faster than propagation. A monoalkylchromium(III) site without an adjacent proton, (≡SiO)2Cr-alkyl, is viable as an active site for polymerization, although its precise origin remains unknown.