提要是有關一書之作者生平及其內容大要、優劣得失的概述，為中國傳統目錄的體制與特色之一。清代乾隆年間修纂的《四庫全書總目》，其提要能承續劉向敘錄之義例，發揮辨章學術、考鏡源流之功用，為後世學者所推崇。四庫提要的撰寫，歷經分纂稿、書前提要、總目提要等階段，由於卷帙浩繁，加上纂修、抄寫、補抄、覆核的歷程前後有二十餘年，以致各階段與各版本之提要間產生差異。有學者主張，四庫提要各種版本的校勘工作，應作為四庫學發展的重點之一，以釐清其間致異之故。本研究旨在比勘現存三種書前提要文字之差異，自四庫著錄圖書中選取三百六十五種，以文淵閣庫本之書前提要為底本，比勘文溯閣與文津閣之書前提要，核其異同。比勘結果，文淵、文溯、文津三閣之書前提要確實存有許多差異，本文略舉若干實例說明。三種書前提要之差異，較常見的是異體字、同義字或文字誤乙（如：冬秋作秋冬，貧賤作賤貧）的情形，可能是因底本不同或抄手的習慣或疏忽所致。而三閣圖書校寫上呈的日期不同，反映的是四庫七閣完成時間不同與長期抄校覆核的事實。在題名、作者、卷數方面，亦有一些出入，有些差異明顯是筆誤或疏忽造成，有些則原因未明，尚待釐清。文字比勘中，最值得注意的是內容的差異，在內容多寡與文字排比方面，文淵閣本與文溯、文津閣本有較多出入。但亦有文淵、文溯閣本相同（似），文津閣本不同；或文淵、文津閣本相同（似），文溯閣本不同；也有三閣內容均有異同的情形。本研究目前只是初步整理出文字的差異，後續有待對內容做進一步分析。

A bibliographic annotation (tiyao提要) is a brief description of the author and content of a book as well as a comment on, or a critique of, the book. The Siku Quanshu Zongmu (四庫全書總目) has long been viewed as a model of the traditional Chinese annotated bibliography and its bibliographic annotations have been praised by many scholars. It is suggested that these annotations can be used as examples for learning how to write bibliographic annotations. The compilation of the Siku Quanshu Zongmu went through three stages: (1) individual draft annotations (分纂稿) written by various scholars, (2) front annotations (書前提要) revised and modified by the officials of the Siku Quanshu Project, and (3) finalized annotations (總目提要) mainly edited and compiled by Ji Yun (紀昀). Initially, the Siku Quanshu had seven written copies and there were seven sets of front annotations. They were housed separately in the seven chambers that Qianlong Emperor (乾隆, r. 1736-1795) built to store the Siku Quanshu. Currently, only three of the seven sets are intact and extant, including Wenyuange (文淵閣), Wensuge (文溯閣), and Wenjinge (文津閣). This study attempts to conduct a collation project of the three versions of front annotations. We chose 365 pieces of front annotations from the aforementioned three sets, respectively. The results corroborate that there exist variations and differences among the three sets of front annotations. This paper presents three examples to illustrate how the collation task was done. Since these annotations were transcribed manually, it is quite common to notice that the three sets might use variant forms for the same character. The descriptions of author, title, or number of volumes might be different as well. In particular, the annotation for the same book might be different slightly or significantly among the three sets. This paper is a summary report of the preliminary findings of the collation task.