This study examined the writing strategies and text characteristics associated with discovering a scientific principle by writing about an experiment. Sixty-four university students (non-science majors) carried out a physics experiment concerning either buoyancy, or the forces acting on a balance scale, then wrote an informal journal-style note about it while thinking-aloud. They provided explanations of the phenomena before experimenting, immediately after experimenting (before writing), and again after writing about the experiment. Students' verbal protocols were segmented and writing operations were identified; the rhetorical structures of their texts were analyzed; and each students' level of general writing strategy was identified. A factor analysis showed that six components accounted for 76% of the variance in these measures. Logistic regression analysis showed that type of science experiment and two writing factors, Problem Solving and Comparisons in Text, discriminated 86% of cases in which students made explanatory gains during the writing interval. The results generally exemplified a metacognitive, problem solving model of writing to learn. However, in contrast to dominant metacognitive models, setting content goals, applying moderately sophisticated writing strategies, and extensive use of content sources were important for learning.