CASE tools provide artifact guidance and process guidance to enhance model quality and reduce their development time. These two types of guidance seem complementary since artifact guidance supports defect detection after each iterative development step, while process guidance supports defect prevention during each such step. But can this intuition be empirically confirmed? We investigated this question by observing developers refactoring a UML model. This study attempted to assess how general were the observations made by Cass and Osterweil on the benefits of guidance to build such model from scratch. It turns out that they do not generalize well: while their experiment observed a benefit on quality and speed with process guidance (but none with artefact guidance), we, in contrast, observed a benefit on quality at the expense of speed with artefact guidance (but none with process guidance).