Perceptual learning improves discrimination of visual features, such as contrast, orientation and spatial frequency, which has been studied intensively in recent years because of its close links to brain plasticity. Past studies for perceptual learning have focused on what we learn to understand the neural representations of learning, while few studies involve information processing during the learning. In this study, we used the multi-stimulus training paradigm to investigate the stimulus coding rules for perceptual learning. We found (1) In contrast with roving, the fixed temporal pattern (fixed stimulus sequence and fixed inter-stimulus intervals) could enable the multi-stimulus perceptual learning. (2) The effect of stimulus roving could not be counteracted by extending the duration of stimulus presentation or presenting physical cues prior to the roving stimulus. On the contrary, the conceptual letter cues associated with stimuli restored learning during the roving. (3) Stimulus roving only disrupted the learning when the memory traces had not been consolidated. Once established learning was resistant, it could be retrieved regardless of the stimulus temporal conditions. Our results suggest that during multi-stimulus perceptual learning, stimulus sequence may tag each stimulus at a conceptual level, so that visual system can switch attention to the appropriate perceptual template during learning.