The life cycle model analyzed here constrains most work on the main job to be full-time. Partial retirement entails a wage reduction and frequently a job change. Maximum likelihood estimates of utility function parameters are designed to incorporate information on the age of leaving full-time work and the age of full retirement. The model closely tracks retirement behavior, including peaks in retirement activity at 62 and 65. The paper explores the relation of these peaks to nonlinearities in the budget constraint created by various retirement programs. Policy analyses based on this and on earlier models with simpler structures are compared.