Geologists and undergraduate students observed eight artificial "rock outcrops" in a realistically scaled field area, and then tried to envision a geological structure that might plausibly be formed by the layered rocks in the set of outcrops. Students were videotaped as they selected which of fourteen 3-D models they thought best represented the geological structure and then explained their choice. The focus of this paper is on how students reasoned from observations to inferences. Students used observations of outcrops' location, steepness (dip), orientation (strike), stratigraphy, and placement relative to topography to infer whether the structure was convex or concave, deep or shallow, symmetrical or asymmetrical, open or closed, and elongate or circular. On average, science majors produced more than twice as many evidence-supported claims than did non-science majors. Science majors produced more valid lines of reasoning than did non-science majors, and students who selected a correct model produced more valid lines of reasoning than students who selected an erroneous model. Apparent challenges included identifying appropriate observational evidence, combining multiple lines of reasoning, and understanding the scale relationship between candidate models and the full-scale structure. (Contains 10 figures, 1 table and 11 notes.)