A job exposure matrix consists of jobs on one axis and substances on the other, with the matrix elements describing the likelihood of an individual's exposure to a substance in a given job. This can be used in case-control studies to infer exposures of subjects whose jobs are known. The simplest form of job exposure matrix contains binary entries, but it is also possible to envisage continuous variables describing the probability of exposure in the job (probabilistic matrix). In such a case, the user has various options for transforming and analyzing the data, including the following: 1) transform to binary variables and analyze as conventional binary exposure variables; 2) leave as continuous variables and analyze using logistic regression; 3) leave as continuous variables and analyze using a linear model. Simulations were carried out to compare the ability of the three methods to estimate odds ratios under 36 experimental conditions. The linear model produced unbiased estimates, the logistic model produced somewhat biased estimates at high odds ratios, and the transformation to a binary variable produced systematically low estimates in most experimental circumstances. With the linear and logistic models, the odds ratio estimators had similar precision when the bias of the latter was not too great. The authors conclude that the linear model permits optimal use of a probabilistic matrix in an epidemiologic study and hope that these results will encourage the development of job exposure matrices containing probabilities rather than dichotomies.