For decades, there has been substantial uncertainty as to the circumstances under which the law will impose liability when the parties have had some negotiations, but had not reached agreement on a fully binding contract, and one of the parties refuses to go further. The law's confusion is partly due to the scholars' failure to recover the law in action governing precontractual liability issues. We show first that no liability attaches for representations made during preliminary negotiations. There is uncertainty when the parties make reliance investments following a "preliminary agreement": that is, they sink costs in the pursuit of a project under an agreement that is too incomplete to enforce, and one of the parties later prefers to exit rather than pursue the contemplated project. Courts have been divided over the question of liability for breach of these preliminary agreements, but a number of modern courts impose on the party wishing to exit a duty to bargain in good faith. Substantial uncertainty remains, however, as to when this duty attaches and what the duty entails. The judicial uncertainty arises, we claim, because key questions have not been satisfactorily answered: Why do parties make such incomplete contracts, then rely before uncertainty is resolved and finally disagree over cost reimbursement when both recognize that their project would be unprofitable? We develop a model which shows that parties create "preliminary agreements" rather than complete contracts when the project they explore could take a number of forms, and the parties are unsure at the outset which form would maximize profits. A preliminary agreement roughly allocates investment tasks between the parties, specifies investment timing and commits the parties only to pursue a profitable project. Parties sink costs in a project because investment accelerates the realization of returns and illuminates whether any of the possible project types would be profitable. A party to a preliminary agreement "breaches" when it delays its investment beyond the time the agreement specifies. Delay will save costs for this party if no project turns out to be profitable and improves this party's bargaining power in the renegotiation to a complete contract if a project would succeed. Delay often disadvantages the promisee, but the main inefficiency is ex ante: When parties anticipate such strategic behavior, the likelihood that they will make preliminary agreements is materially reduced. This is unfortunate because the performance of a preliminary agreement often is a necessary condition to the creation of a complete contract and the subsequent realization of a socially efficient opportunity. Thus, contract law should encourage relation-specific investment by awarding verifiable reliance costs to a party to a preliminary agreement if its partner has strategically delayed investment. We study a large sample of appellate cases that deal with reliance prior to the signing of a complete contract. This study reveals that (a) parties appear to make the preliminary agreements we describe and breach for the reasons our model identifies; and (b) courts sometimes protect the disappointed party's reliance interest when they should, but the courts' imperfect understanding of the parties' behavior leads them to err.