We test hypotheses about glacial dynamics by evaluating the ability of a linear statistical model to simulate climate during the previous ~800,000 years. During this period, the linear model simulates the timing and magnitude of glacial cycles, including the saw-tooth pattern in which ice accumulates gradually and ablates rapidly, without falsely simulating an interglacial after each peak in obliquity. Conversely, the linear model fails to simulate experimental observations that are created by a nonlinear data generating process. Together, these (in)abilities suggest that nonlinearities, threshold effects, bifurcations, and/or phase-specific governing equations do not play a critical role in glacial cycles during the late Pleistocene. Furthermore, the model’s accuracy throughout the sample period suggests that changes in orbital geometry create the Mid-Brunhes event.