The considerations and procedures which shape the contractual arrangements for the procurement of reliable aerospace systems are discussed. Such systems are usually procured from single sources of supply on a relatively rigid schedule, and they consist of high-cost, low-production, long-lead-time items. Furthermore, cost of maintenance over equipment lifetime may exceed the initial cost by a large factor; hence, logistics considerations also exert a strong influence. In these special circumstances, normal commercial procurement procedures are unsuitable for providing a customer with the protection inherent in the competitive market place, and he must resort to policing his supplier's activities, which at best is never a completely adequate procedure. The contract must provide for a program plan which establishes specific scheduled monitoring points during the development and fabrication stages, such as review of design requirements and environments, design arrangements, proof-of-design by analysis, qualification test plans and execution, and acceptance test procedures. In the execution of this plan the customer assumes the risk that a product which passes the qualification test will be satisfactory in service. The supplier assumes the risk that he can design and fabricate a product which will pass the qualification test. Qualification test results should therefore correlate both with analytical design methods and with data fed back from actual field operations. The proper design of qualification test procedures depends on an understanding of the basic causes of unreliability, such as our inability to predict correctly the actual operating environments, and the unit-to-unit variation in component resistance to failure. Our inability to specify correctly the environmental requirements is due in part to the inadequacy of available empirical data; to new and previously unknown environments encountered as the operating range is extended; and to the invention by the operators of new uses for the system. The considerations which enter into the design of qualification test spectra are discussed in detail, and some spectra in current use are analyzed. The unit-to-unit variation of components may be minimized by reliability control of the design of the manufacturing processes, and may be measured in a series of tests-to-failure. Although it is difficult to establish accept-reject criteria, an approach is described which is believed to have merit.