While virtual reality (VR) training has been used in many examples of specialized skills training, its learning effect has also been explored with a focus on gross motor performance. Using VR can develop technical students' procedural knowledge in a safe environment while also saving the real materials, which is an essential issue of sustainability. However, the procedural accuracy and performance quality of learning outcomes of practicing VR have not been extensively examined. To address this gap, the present study designed VR for car detailing to explore how two types of learning outcomes could be predicted by students' self-efficacy that drives their learning interest and situational anxiety mediated by flow experience while practicing with VR. Based on a single group quasi-experiment study, the learning outcome tests were administered in the first, second, and fifth weeks after four rounds of practice to 143 technical students who had undergone VR car detailing training. Data were subjected to confirmatory factor analysis with structural equation modeling. Results revealed that when VR was used for car-detailing training, the learning outcomes of procedural accuracy and performance quality could be predicted by participants' flow experience. Learners' flow experience could be positively predicted by VR learning interest and VR using anxiety. Lastly, VR self-efficacy was positively related to VR learning interest, but negatively related to VR using anxiety. It is expected that the results can be applied to other technical training designs.