Method effects can be additive (independent of trait correlations) or multiplicative (associated with trait correlations). This study is the first to empirically assess the relationship between the nature of method effects and the goodness-of-fit of different latent factor models. Specifically, we examined method effects in 17 published multitrait-multimethod data sets and evaluated the usefulness of confirmatory factor analysis, the direct product approach, and Marsh's correlated uniqueness technique for modeling these effects. While each of the models fit some of the data sets well, Marsh's technique appeared to be generally more effective. Also, Campbell and O'Connell's slope index indicated that additive models (confirmatory factor analysis and the correlated uniqueness approach) were not more likely than a multiplicative model (the direct product model) to provide a good fit to data with additive method effects; nor did a multiplicative model provide a better fit than additive models when method effects were multiplicative.