The standard method of estimating the value of travel time variability for use in policy appraisal is to estimate the parameters of a reduced-form utility function, where some measure of travel time variability (such as the standard deviation) is included. A problem with this approach is that the obtained valuation will in general depend on the standardized travel time distribution, and hence cannot be transferred from one context to another. A recently suggested remedy of this problem has been to estimate a scheduling model, which in theory is transferrable, and use the implied reduced-form to derive valuations for use in appraisal. In this paper we estimate both a scheduling model and the implied reduced-form model, using stated choice data. The valuation of travel time variability implied by the scheduling model turns out to be substantially smaller than what is obtained from a reduced-form model estimated on the same sample. The results suggest that the scheduling model does not capture all of the disutility arising from travel time variability. Hence, although it can be shown that scheduling and reduced-form models are “theoretically equivalent”, that hypothesized equivalence is not reflected in the empirical evidence. We speculate that the derivation of reduced-form models from an underlying scheduling model omits two essential features: first, the notion of an exogenously fixed “preferred arrival time” neglects the fact that most activities can be rescheduled given full information about the travel times in advance, and second, disutility may be derived from uncertainty as such, in the form of anxiety, decisions costs or costs for having contingency plans. We also report our estimates of the valuation of travel time variability for public transit trips, for use in applied appraisal.