We use data on sequential water auctions to estimate demand when units are complements or substitutes. A sequential English auction model determines the estimating structural equations. When units are complements, one bidder wins all units by paying a high price for the first unit, thus deterring others from bidding on subsequent units. When units are substitutes, different bidders win the units with positive probability, paying prices similar in magnitude, even when the same bidder wins all units. We recover individual demand consistent with this stark pattern of outcomes and confirm it is not collusive, but consistent with non-cooperative behavior. Demand estimates are biased if one ignores these features.