The purpose of this paper is to determine whether predictor variables, selected from readability research and from teacher nomination, might be used to grade children’s early reading books. Two hundred Reading Recovery books were rated on five variables thought to represent ‘semantic’, ‘syntactic’ and ‘phonic’ features of text. The variables accounted for 83% of book level variance. In a cross-validation exercise they were found to be stable and reliable. It is suggested that the data support the validity of the Reading Recovery book levels, and that the formula can be used by teachers to grade children’s early reading books.