Student construction of models is a strong focus of current research and practice in science education. In order to study in detail the interactions between students' model generation and evaluation and their development of explanatory ideas to account for magnetic phenomena, a multi-session teaching experiment was conducted with a small number of fifth grade students. Two small groups received full scaffolding, and two small groups received partial scaffolding. Students in both fully and partially scaffolded groups were asked to make their own predictions, observations, and explanations for scientific phenomena. Then they were asked to elaborate individual ideas and discuss with others in order to select or develop the best group consensus model and to compare their current group model with their previous group models. Only fully scaffolded groups were explicitly asked to consider the scientific modeling criteria of visualization and explanatory power. Through reflection on their explanations using these modeling criteria, most students in the fully scaffolded groups gradually developed, evaluated, and revised their explanations to coherent and sophisticated explanatory models. By contrast, none of the students in the partially scaffolded groups, who relied only on self-generated model evaluation criteria, revised their original fragmented ideas toward more coherent and sophisticated explanations. Through a detailed analysis of the evolution of the students' thinking in both fully and partially scaffolded groups, we explore the ways in which the explicit scaffolding of the scientific modeling criteria seemed to aid in the evolution of their ideas toward more sophisticated and coherent explanatory models.