Empirical search cost estimates are often large and increasing in the size of the transaction, even if search can be done conveniently online. To assess this pattern systematically, we conduct an online search experiment in which we manipulate the price scale while keeping the physical search effort for each price quote constant. We also record the time subjects need to obtain a price quote in order to derive a direct measure of subjects' time and hassle costs of search. Based on a standard search model, we confirm that search cost estimates are large relative to directly elicited search costs and increasing in the price scale. We then modify the search model to allow for context effects, i.e., the tendency that people become less sensitive to price variations of fixed size when the price scale or range of outcomes increases. With the modified model, we find scale-independent search cost estimates that correspond well to subjects' directly elicited search costs. We show that the consumer welfare losses from context effects can be quite substantial and discuss how empirical work could deal with scale-effects.