We report an experiment which seeks to determine how novice users' conceptual understanding of graphs differs depending on the nature of the interaction with them. Undergraduate psychology students interpreted three-variable "interaction" data in either bar or line graph form and were required to either think aloud while doing so or to produce written interpretations. Analysis of the verbal protocols and written interpretations showed that producing a written interpretation revealed significantly higher levels of comprehension than interpreting them while thinking aloud. Specifically, a significant proportion of line graph users in the verbal protocol condition was either unable to interpret the graphs, or misinterpreted information presented. The occurrence of these errors was substantially lower for the bar graph users in the verbal protocol condition. In contrast, analysis of the written condition revealed no significant difference in the level of comprehension between the two graph types. Possible explanations for these findings are discussed.