Land can be inefficiently allocated when attempts to assemble separately-owned pieces of land into large parcels are frustrated by holdout landowners. The existing land-assembly institution of eminent domain can be used neither to gauge efficiency nor to determine how to compensate displaced owners adequately. We take a mechanism-design approach to the assembly problem, formalizing it as a multilateral trade environment with perfectly complementary goods. We characterize the least-inefficient direct mechanism that is incentive compatible, self-financing, protects the property-rights of participants, and does not assume that participants have useful information about the subjective valuations of others. The second-best mechanism, which we call the Strong Pareto (SP) mechanism, utilizes a second-price auction among interested buyers, with a reserve sufficient to compensate fully all potential sellers, who are paid according to fixed and exhaustive shares of the winning buyer's offer. It may also internalize local externalities. While the SP mechanism only approves efficient sales, efficiency is not sufficient for sale---even with competitive bidding---because the auction reserve may exceed the aggregate seller valuation. The inefficiency of the second-best mechanism implies a Myserson-Satterthwaite (1981)-style impossibility theorem. We propose a criterion that encompasses concern for both efficiency and the rights of property owners to evaluate the relative performance of assembly mechanisms and the efficiency cost of strict adherence to individual rationality. In a simple example, we compare the expected outcome of the SP mechanism with two alternatives: a plurality mechanism based on SP, but with a lower reserve that is only high enough to fully compensate a plurality of owners and a stylized model of eminent domain.