In this paper, we present an analysis of the Arkansas Teacher Retirement System (ATRS) pension plan and an empirical investigation of the behavioral response to that plan, as well as to a possible reform plan. We begin by describing the plan parameters and discussing the incentives these parameters create. We then estimate the effect of pension wealth accrual on teacher separation decisions using a new longitudinal dataset of Arkansas teachers. The resulting coefficients are subsequently used to compare predicted separation probabilities under the current parameter regime versus a constant accrual retirement plan. We find evidence that teachers' retirement decisions are sensitive to pension wealth accrual. Our simulations suggest that a constant accrual plan could notably smooth the pattern of retirement behavior from that currently observed. (Contains 2 tables, 16 figures, and 22 footnotes.)