With increasing use of the World Wide Web, rapid scanning or skimming of material on screen has become a frequent activity. However, the outcome of this method of reading has not been thoroughly investigated. Using a range of question types, comprehension was measured after reading from screen at both a normal and fast reading speed. In addition, by automatically recording how readers scrolled through each document, reading patterns were explored. A speed-accuracy trade-off was found and, in general, the recall of specific details was less accurate than responses to ‘higher order’ questions. However, questions that addressed the structure of the text were hardest. Analysis of the scrolling movements showed that the overall time spent pausing between movements was the best predictor of comprehension. At a normal reading speed, the most effective readers, in terms of higher comprehension scores, were those who spent more time between scrolling movements, which were fast and frequent.