In an inefficient visual search task, when some distractors (old items) temporally precede some others (new items), the old items are excluded from the search, a phenomenon termed visual marking. Past studies have shown that this effect is unaffected by luminance change of old items but is eliminated by their shape change, suggesting that the shape identity has to be maintained for successful visual marking. It was unclear either the chance in meaning or shape was critical, because the shape changes accompanied meaning changes in the previous studies. The present study examined whether the consistency in the semantic or only graphical identity of old items was critical for visual marking. Under the meaning-change condition, on every trial each of the old items (pictures of butterflies) was replaced with a Japanese Kanji character representing “cow”, creating both semantic and graphical changes. Upon these changes, new items (Kanji characters representing “dog”) appeared at blank locations. Under the meaning-change condition, the old items (pictures of cows) turned into Kanji characters of “cow”, creating graphical changes. Participants searched for a tilted character of “dog”. The results indicated that the search under the meaning-unchanged condition was more efficient than that under the control condition in which all items appeared simultaneously. Under the meaning-change condition, however, the search efficiency did not improve relative to the control condition. These results indicated that visual marking survived the graphical change of old items as long as their meaning was maintained. A follow-up experiment replicated the preserved visual marking over graphical change and excluded an alternative that the results of the main experiment were due to an artifact of used pictures. These results suggested that consistent semantic identity during the preview period is sufficient to reveal visual marking even when graphical identity discontinued.