The design of many engineered structures is governed by considerations of material fatigue failure. For example, offshore structures must withstand continuous oscillatory wind and wave loads. Under extreme conditions, the loading may fluctuate significantly over short time periods. Substantial damage, and most failures, are caused by these short-time loads. Existing methodologies for evaluating the probability of fatigue failure assume that the loading process is stationary (that is, the loading characteristics do not change with time), which may not be appropriate for extreme loading conditions. This study develops a reliability model for fatigue failure that accommodates nonstationary loading processes. The model describes the response of the structure to nonstationary loads as a random process. The crossing statistics of this response process are calculated in order to determine the expected number of load reversals; the accumulated damage resulting from these load reversals is determined using the methods of fracture mechanics. Finally, the estimated damage can be compared to a limiting criteria so as to calculate the probability of failure. This methodology improves upon existing procedures for the design of structures prone to fatigue failure under nonstationary loads, such as those caused by an earthquake. This information can also be used to specify rational design, inspection, maintenance, and repair schemes for many types of structures.