A study examined whether conclusions about constructive processes in reading based on analysis of group data were consistent with those based on an analysis of individual data. Subjects, selected from a larger sample of 45 first grade students who had participated in a longitudinal study on acquisition of linguistic procedures for printed words, were from a variety of ethnic and cultural backgrounds and were relatively homogeneous in socioeconomic indicators. The same data set of reading-related processing measures was analyzed in four different ways. In the first approach, the analysis was based on achievement groups, aggregated over children drawn from different classrooms, with variation among children within the group treated as error. In the second approach, the analysis was based on instructional groups taught by the same teacher, with error estimated as in the first approach. Analysis in the third approach was based on the same children as the second approach but variation among children was treated as systematic variance and individual responses over stimulus trials were used to estimate error. Finally, in the fourth approach, separate analyses were performed for each child in the instructional groups in a design that used variation over stimulus trials to estimate error. Results depended on how the data were aggregated: the first two analyses indicated that there is one process in reading acquisition, while results of the last two analyses indicated that there is not one single process in learning to read. Rather, even when the instructional program was held constant, variation in children's constructive processes resulted in the learners using instructional cues in more than one way. Differences in conclusions were attributed to the ecological fallacy in which inferences about one unit of analysis are based on analyses at another level. (Five tables of data are included; 60 references are attached.) (Author/RS)