The purposes of the present investigation were to examine the dimensionality of the Rotter Internal-External (I-E) scale, to test the invariance of its factor structure before and after an intervention designed to alter I-E, and to demonstrate the use of confirmatory factor analysis. An a priori model based on the content of the Rotter items hypothesized six factors, and a review of 20 earlier factor analyses demonstrated that each of these factors had been previously identified. Results from time one and time two both indicated that this hypothesized model was able to adequately explain responses to the Rotter items, that the hypothesized factors were defined and distinguishable, that the factor structure was invariant over time, and that alternative models with fewer hypothesized factors or a simpler structure did not fit the data as well. In a more general discussion of the Rotter scale it was concluded that, when judged by current standards, the Rotter I-E scale is a poor measurement instrument: its internal consistency reliability is minimal, the rationale for its forced-choice format is dubious, the assumption of its unidimensionality is clearly wrong, and the distinguishable facets of the scale cannot be inferred with sufficient reliability to be practically useful. (Author)