We introduce a novel and general method for how to make choices without assuming common knowledge or equilibrium behavior. We apply this method to bidding in first-price auctions and test our recommendations using data on bidding in auctions. The method prescribes to first rule out some environments and then to find a so-called robust (bidding) rule that performs well in each of those remaining. This method generates explicit bidding recommendations that we test in laboratory experimental auctions as well as in field data which come from timber auctions. Our recommendations outperform the bids made in the data.