Based on a review of research and a cognitive devleopment model (Bigg & Collis, 1991), we formulated a framework for characterizing elementary children's statistical thinking and refined it through a validation process. The 4 constructs in this framework were describing, organizing, representing, and analyzing and interpreting data. For each construct, we hypothesized 4 thinking levels, which represnt a continuum from idiosyncratic to analytic reasoning. We developed statistical thinking descriptors for each level and construct and used these to design an interview protocol. We refined and validated the framework using data from portocols of 20 target students in Grades 1 through 5. Results of the study confirm that children's statistical thinking can be described according to the 4 framework levels and that the framework provides a coherent picture of children's thinking, in that 80% of them exhibited thinking that was stable on at lest 3 constructs. The framework contributes domain-specific theory fo characterising chilren't statistical thinding and for planning instruction in data handling.