In order to mitigate the high environmental impact of the construction sector, the Dutch government envisions a fully circular building economy by 2050. The first phase of this transition (2018-2023) consists of experimentation. Here, circular building experiments should trigger radically new ways of thinking and radically new ways of doing. Learning-by-doing is the device, but clear conceptual knowledge on how to achieve deep learning is lacking. Therefore, this explorative research aims to assess how circular building experiments could be designed in order to stimulate deep learning. Stakeholders involved in the design and construction process of four circular building experiments in the Netherlands have been research. Based on transition and organizational learning literature, for each case the subjects (who learns), the process (learns how), the objects (learns what) and the effect (outcome) have been assessed. The case-studies show that deep learning is achieved in every experiment and that it has changed organizations routines, providing evidence that experiments are an effective means to trigger radical change. However, the potential of transition learning is higher, given the fact that the learning activities in the experiments and structures in the organizations involved were missing or inadequately approached. Therefore, this research recommends to more carefully design experiments to support transition learning. In doing so, this research proposes a typology for transition learning based on seven flavours, which allow researchers to understand, set-up and evaluate transition learning in experiments. Here, combinations of flavours can yield different outcomes. Furthermore, for practitioners a roadmap with checklist questions has been developed to support setting-up future circular building experiments.