This study investigates children's semantic integration of sentence information as a function of instructions (form or substance), test sentence form (verbatim or paraphrased from acquisition story sentences), and story content (spatial or general relationships). After 144 fifth-grade children were presented with twelve short acquisition stories, they were asked to discriminate between true and false premise or inference recognition test sentences. Performance varied as a function of experimental condition, and discrimination for correct premises was much better than that for correct inferences. No consistent support was found for a view of constructive memory in children. (Author)