The purpose of this study was to investigate the effect of analogical learning on the acquisition of falsification reasoning skill. Two hundred and twenty-seven college students were randomly assigned to three analogical learning situations (worked-out examples, worked-out examples + logical structure illustration, control) to work on four practice items with or without falsification worked-out examples. Their performance on 16 test items of deontic rule testing showed that worked-out examples could enhance the use of falsification reasoning only when test items shared the same semantic structure with the worked-out examples. When test items differed from the worked-out examples in the semantic structure, the success of analogical learning occurred only when there was no surface similarity between them. These findings suggest that in scientific reasoning a scientist’s challenge lies in extracting a problem’s underlying logic of rules and meanwhile excluding the misleading effect of surface similarity.