A method was developed for characterizing and comparing the semantic content of picture-text combinations used as stimuli in research on reading procedural instructions. The written directions for assembling a toy model were analyzed, and sets of propositions in the instructions were constructed using C. H. Frederiksen's discourse analysis model. Using the same model, written descriptions of pictures were consolidated and then reduced to sets of interrelated propositions. A comparison of the two sets resulted in revisions of text and pictures to obtain redundancy of essential information. Three general types of information were identified as "core" components of both forms: locative, referential, and sequential information. Stimulus materials attempting to emphasize each type of information in both text and pictures have been prepared and are being tested. So far the results of such testing have produced the following observations: (1) the linear nature of text makes core sequential information inherent, so manipulation of sequential information is difficult and may destroy the integrity of the instructions; and (2) core locative information is inherent in pictures and cannot be manipulated without changing the essence of the messages. (RL)