Studies comparing virtual and physical manipulative environments (VME and PME) in inquiry-based science learning have mostly focused on students' learning outcomes but not on the actual processes they engage in during the learning activities. In this paper, we examined experimentation strategies in an inquiry activity and their relation to conceptual learning outcomes. We assigned college students to either use VME or PME for a goal directed physics inquiry task on mass-spring systems. Our analysis showed that the best predictors of learning outcomes were experimental manipulations that followed a control of variable (CV) strategy, with a delay between manipulations ("systematic inquiry"). Cluster analysis of the prevalence of these manipulations per participant revealed two distinct clusters of participants, systematic inquiry or not. The systematic inquiry cluster had significantly higher learning outcomes than the less systematic one. Furthermore, the majority of the participants using the PME belonged to the more systematic cluster, while most of the participants using the VME fell into the non-systematic cluster, likely because of the specific affordances of the real and virtual equipment they were using. However, beyond this impact on inquiry process, condition had little effect. In light of these results, we argue that investigating processes displayed during learning activities, in addition to outcomes, enables us to properly evaluate the strengths and weaknesses of different learning environments for inquiry-based learning. [For complete proceedings, see ED560503.]