We explore improving support for revisitation in documents by automatically generating bookmarks based on users’ reading history. After showing that dwell time and number of visits are not appropriate for predicting revisitations in documents, we model the high-level reading task as a sequence of reading blocks and recognize long-distance scrolls as separators between them. A long-distance scroll is defined as a continuous scrolling action which causes the document to be navigated beyond a one-page distance. We propose a new technique, called the Head-Tail (HT) algorithm, to generate bookmarks at the head and the tail of reading blocks, whose validity is quantitatively verified by log data analysis. Two studies were conducted to investigate this HT implicit bookmarking technique. The first is a controlled experiment that compared the HT algorithm to the widely used simple recency algorithm for generating implicit bookmarks, in terms of revisit coverage ability and distance between bookmarks. Results showed the HT algorithm to be superior in both measures. The second is a more ecologically valid study that investigated implicit bookmarking performance in real reading tasks, using Adobe Reader integrated with our implicit bookmarking technique. Results showed that our technique covered 85.1% of revisitations and saved users from 66.0% of long-distance scrolling actions. We end with a discussion of how to encourage users to use implicit bookmarks.