Confidence Weighting (CW), after Ebel, and Multiple Responding (MR), after Coombs, are compared empirically to determine which improved test reliability more in the case of a course pretest derived from the final examination. It was hypothesized that MR, which purportedly measures partial knowledge, would be more effective than CW, which measures extra knowledge. The subjects were 58 students enrolled in a graduate course in Adolescent Psychology. The pretest comprised 20 four-choice items drawn from the final exam. One randomly selected half of the class took the pretest under CW instructions; the other, under MR instructions modified to yield a "raw" score also. The reliability of raw scores was zero in both groups. MR-score reliability improved to .38 while CW-score reliability remained zero. (Author)