This study used many-faceted Rasch measurement to investigate the difficulty of graded readers using a 3-item survey. Book difficulty was compared with Kyoto Level, Yomiyasusa Level, Lexile Level, book length, mean sentence length, and mean word frequency. Word frequency and Kyoto Level were found to be ineffective in predicting students' perceptions of book difficulty. Book length was found to be highly predictive of perceived book difficulty, with the Yomiyasusa Levels predicting 68% of variance, while the Lexile measure of mean sentence length was moderately predictive, with 40% of variance explained. These results show that current headword levelling of graded readers is ineffective and that publishers' book levels do not provide useful guidance in selection of books to read. It is therefore recommended that students use book length as their primary consideration in choosing books and that reading recommendations and purchasing decisions be based on Yomiyasusa Levels rather than publishers' levels.