A readability survey was carried out on a wide selection of history books, using a computer program, ASTRA 3, developed for the study. The results showed that although the mean readability levels of textbooks provided for 12-14 year olds were appropriate to their readership, there were very large and seemingly random internal fluctuations. This, coupled with an observed lack of scalar correspondence between formulae, has implications for current practices in readability assessment. The responses of readers to readability fluctuations were next investigated, using one subjective measure and three behavioural measures, namely, children's subjective ratings, cloze procedure, reading rate, and stress reaction (Skin Conductance Response). Relatively low cost microcomputer based apparatus was developed for studying the latter two measures. It was found that average 12-14 year old readers were able to perceive variation in difficulty in adjacent passages and these subjective ratings proved to be relatively good predictors of fluctuations in cloze procedure scores. A measure of redundancy based on cloze responses was also shown to be strongly related to children's subjective ratings of contextualised passages. No statistical relationship was found, however, between textual cohesion and either the subjective or the objective indices. These findings raise a number of questions of concern to writers and publishers of school text books. According to the cloze test, all the passages were at frustration level, although this is contrary to what would have been expected from the range of readability indices. Reading rate tended to increase, rather than decrease with difficulty, though in absolute terms the variation was slight. SCR responses tended to be labile and snowed substantial individual. differences. This part of the study opens up a number of questions about the appropriateness of different ways of examining the response of readers to specific 'text features that contribute to readability. In general, the findings in this interrelated series of studies show that a great deal of readability assessment is based on assumptions of dubious validity concerning the distribution of difficulty within texts, the nature of relationships between different readability measures, and the ways in which readers actually respond to hypothesized difficulties.