Similarity promotes the integration of visual information and reduces the load of representations in working memory during the encoding stage, and improves the visual working memory performance. Researchers have found that spatial visualizers’ performance decreased evidently from short delay to long delay in a high-load condition, while object visualizers performed stably. This advantage results from the higher neural efficiency of object visualizers than spatial visualizer during the retention stage rather than encoding stage in visual working memory. However, whether similarity affects the representations of visual working memory for object visualizers and spatial visualizers is still unclear. The purpose of the present study is to investigate how similarity modulates the performance of object and spatial visualizers and the mechanism of object and spatial visualizers’ representation in visual working memory. In the current study, we used Object-Spatial Imagery Questionnaire –Revised to categorize participants into object visualizers and spatial visualizers. The experimental materials were squares with similar or dissimilar colors. Utilizing the change detection paradigm, we designed two experiments. In experiment 1, we adopted 2(cognitive style: object visualizer, spatial visualizer) × 2(similarity: similar, dissimilar) × 2(delay: 1000ms, 3000ms) mixed design. In experiment 2, we adopted 2(cognitive style: object visualizer, spatial visualizer) × 2(similarity: similar, dissimilar) mixed design and the delay was 3000ms. Cognitive style was a between-subject factor, and the other variables were within-subject factors. The accuracy rate was recorded in two experiments, and the event related potential (ERP) was recorded in experiment 2 during the retention stage of working memory. The behavioral results showed that the accuracy in similar condition was significantly higher than that in the dissimilar condition, as well as that in short delay condition than long delay condition. Most importantly, a significant interaction among delay, similarity and cognitive style was observed. In the similar condition, there was no obvious differences between object visualizers and spatial visualizers in both short and long delay conditions. In the dissimilar condition, there was no difference between object visualizers and spatial visualizers in the short delay condition, whereas object visualizers have higher accuracy than spatial visualizers in the long delay condition. The result suggested that object visualizers performed more stable than spatial visualizers in maintaining dissimilar objects as the time went by. The ERP results showed that larger negative slow cortical potentials were evoked for spatial visualizers than object visualizers in the dissimilar condition rather than the similar condition, which suggested that spatial visualizers spent more cognitive resources than object visualizers at the retention stage of working memory in the dissimilar condition. Taken together, these findings have revealed that the object visualizers have advantages while processing the dissimilar objects instead of the similar objects in visual working memory compared to the spatial visualizers. The underlying mechanism is that object visualizers have higher neural efficiency than spatial visualizers during the retention stage of dissimilar objects in visual working memory.