A common approach to the study of visually guided action is to search for optical invariants that reliably guide action across a range of conditions. However, optical invariants are not always available, and the reliability of non-invariants can change with local constraints. So the visual system must be able to flexibly tune to different optical variables. We investigated such flexibility within the context of a simulated braking task. In a previous study, observers braking before a stationary target relied on global optic flow rate (GOFR). When braking behind a moving lead target, GOFR is unreliable because it specifies the observer's absolute speed rather than observer-target relative speed. We predicted that participants could learn to disregard unreliable GOFR cues, but only to the extent that other reliable information is available. Participants used a footbrake to decelerate from a rapid approach and make soft contact with a moving lead target. Observer absolute speed and observer-target relative speed were independently manipulated, and target size was either fixed or variable. A combination of optical angle and expansion rate closely correlates with the required deceleration when target size is fixed, but not when it varies. Participants in the variable-size condition exhibited a weak GOFR bias, but those in the fixed-size condition were unaffected by GOFR, apparently because other reliable information was available that allowed them to ignore GOFR. The results highlight the robust process of flexible attunement, and suggest that the single optical invariant assumption is too rigid to account for successful visually guided action.