We used delayed sentence-picture verification tasks to investigate multilingual perceptual representations. In experiment 1, participants listened to sentences with implied shapes. After a 10-min interval, they judged whether pictures had been mentioned in the preceding sentences or not. Results in experiment 1 showed significant match effect in L1, but not in high proficient L2 or low proficient L3. In experiment 2, Participants listened to one language block, then immediately judged one picture block, totally three language-picture blocks. Results in experiment 2 were parallel to results in experiment 1. Our study supports the view of distributed conception: L2 and L3 are associated with less perceptual symbols than L1, indicating great impact of acquisition styles on perceptual representations. Our results show little impact of language proficiency levels on perceptual representations in delayed tasks.