This paper examines the behavioral and stochastic aspects of modeling emission reductions from vehicle Inspection and Maintenance (I/M) programs. Forecasts of the potential emission reductions from such programs have been modeled by the use of the Environmental Protection Agency's MOBILE Model, EPA's computer model for estimating emission factors for mobile sources. We examine the structure of this Model and review the way behavior of drivers, mechanics and state regulatory authorities is incorporated in the current generation of the Model. We focus particularly on assumptions about vehicle repair under I/M, compliance with I/M requirements, and the impact of test measurement error on predicted I/M effectiveness. We also include some preliminary comparisons of the Model's outcomes to results of the I/M program in place in Arizona. Finally, we perform some sensitivity analyses to determine the most influential underlying parameters of the Model. We find that many of the assumptions of the I/M component of the Model are based on relatively small data sets on vehicle done in a laboratory setting, and that the output of the Model makes it difficult to compare the results against real world data from on-going state programs. In addition, the Model assumes that vehicles will either be repaired or receive a waiver. In the Arizona program there appears to be a third category of vehicles -- those which fail the test and do not receive passes. This share may be as high as a third of all failing vehicles. Vehicles which do not eventually pass the test would be treated in the Model as non-compliant. However, in current programs, states do not seem to be measuring and entering the compliance rate correctly. The paper also examines the evidence about whether emissions deteriorate over the life of vehicle in a grams per mile basis (as assumed by the Model) or a grams per gallon basis. It finds support for the argument that emissions deteriorate on a grams per gallon basis. We find through sensitivity analysis that the repair effectiveness assumed by the Model to occur in an IM240 test are much greater than for the idle test, and that identification rates and repair effectiveness vary a great deal according to the cutpoint. These results are based on small numbers of vehicle tests in a laboratory setting and could be compared to real world evidence. Examining costs and cost-effectiveness of variations in I/M programs is important for determining improvements in I/M programs. States may not have incentives to develop cost-effective programs based on current Model that forecast emission reduction "credits" that are overly optimistic.