Confirmatory factor analysis (LISREL VI) is the method best suited to the comparison of measurement models when those models are based on a priori assumptions. Traditionally, positive and negative item stems were mixed on affective scales to reduce response set bias since the item pairs were considered to be parallel. Recent studies indicate that positive and negative item stems may form separate factors, implying that they represent different constructs. In this study, the differences between positive and negative item stems were assessed using two forms of a computer anxiety scale to ascertain if the negation of an item produces a parallel item and to compare the factor structures and measurement errors to determine if factor invariance can be claimed. Three forms (Forms A, B, and C) of a computer anxiety scale were administered to a random sample of students (20 homerooms) at a small city high school. Reverse scoring was used for all items on Form B and for appropriate items on Form C. The results are consistent with those of other researchers, providing more evidence that the use of reverse scored items on an affective scale can alter students' responses to an item. One should view results with caution when the instrument includes mixed item stems, since the negation of an item tends to lead to an increase in the error variance related to the item. In general, positive and negative forms of this scale do not meet the criteria for factor invariance or for parallel tests. (TJH)