The study compared three methods of instruction in binary and octal arithmetic, i.e., (1) Norman Crowder's branched programed text, "The Arithmetic of Computers," (2) another version of this text modified so that subjects could not see the instructional material while answering "branching" questions, and (3) a narrative text version presenting the same content material. The principal behavioral measure was relative performance on a pre- and post-trainin criterion test. The results indicated that prohibiting visual contact with instructional material while answering questions significantly increased the number of erroneous alternatives selected by the subjects, but did not significantly alter the amount of learning manifested nor the time necessary to complete training. The programed instructional methods resulted in significantly greater improvement on the criterion test than was attained by using the narrative text. The time to complete instruction was significantly less with the narrative text version of the material. Records were kept of the number of "wrong answer" branches taken by the subjects receiving instruction via the branched programs. Only about 6% of the total possible "wrong" branches were actually taken. This suggests that branched programing may be wasteful by virtue of providing a quantity of material that is never studied. (DDC)