AbstractAn earlier analysis of the inspection and maintenance (I/M) program in Phoenix, Arizona [Environ. Sci. Policy 4 (2001) 377] found that fleet average emissions increased or “deteriorated” substantially between biennial I/M test cycles, and that a large fraction of the vehicles that initially failed and later passed in the first I/M cycle failed again two years later. In this paper we study vehicles that reported for a second I/M cycle well before their required biennial cycle, often because the subject vehicle was about to be sold to a new owner, in the decentralized California ASM program and the centralized Phoenix IM240 program. Because the off-cycle tests occurred at different times after the initial cycle, they allow the analysis of repeat failure rates over the shorter term (less than two years). We found that in California 20% of fail–pass vehicles, and 5% of initial-pass vehicles, failed the initial test of their next I/M cycle occurring within three months of completing the previous cycle; the failure rates were even higher in the centralized Phoenix program. This suggests that highly variable emissions, and not test fraud, were the cause of the large number of vehicles failing so soon after completing a previous I/M cycle. According to a model, as much as 74% of the vehicles that failed their initial test in Phoenix passed a retest because of emissions variability, and not because any repairs were performed. Vehicles with highly variable emissions may not only fail and pass a subsequent retest without repair, but also are likely to fail again if tested shortly thereafter. A large number of these vehicles may explain the apparent rapid increase or “deterioration” in fleet emissions observed in multiple years of I/M test result data.