I have been running Web usability studies since 1994 and the most striking conclusion from looking back is that most findings about Web usability are the same now as they were in 1994. This may be surprising, but usability is about basic human capabilities and users' needs which do not change nearly as rapidly as technology.
The dominant Web look in 1994 was a mixture of gray text and large images, as shown to the right (screens from a comparative study of sites in 1994). This has certainly changed: We have better layout capabilities and many designers have learned to minimize their use of graphics and save download time (though still not enough). Despite the change in looks, many of the findings from 1994 continue to hold:
Users don't read on the Web: They scan the text.
Some amount of personality (the "author's voice") makes sites more attractive: users don't like bland impersonal corporate sites.
Web users are impatient: They want to get their answers immediately and do not want to be slowed down by "cool" features, mission statements, or self-promoting grandstanding.
Users often print out pages: They don't trust the site to have the page for them if they need it at a later date (and they still don't trust sites to be stable: a rather sad finding).
Download times are becoming ever more critical and sites need to design for speed. Users have always requested fast pages, but in the early years, a novelty factor made users slightly more tolerant of slow downloads. This tolerance has declined markedly in recent years.
Search was always liked by users, and has now become mandatory for any large site since the amount of content keeps growing.
There are some new findings due to technology introduced after 1994:
Animation is almost always annoying and should be avoided most of the time
Applets should sometimes open their own window and leave the browser behind
Wild background patterns disrupt users' reading; use colored text with care and avoid blue for non-link text (and retain blue as the standard link color for unvisited links)
There are also a few cases where early findings have to be modified, as described in the following.
Scrolling Now Allowed
In early studies, I found that only 10% of Web users would scroll a navigation page to see any links that were not visible in the initial display. The vast majority of users would make their selection from those links they could see without scrolling. In retrospect, I believe this was due to people treating a set of Web options like they would treat a dialog box: You always design dialog boxes so that all choices are visible (except for tabbed dialogs which are known to have severe usability problems; and the tabs do indicate the amount and nature of the hidden options).
In more recent studies, we have seen that most users scroll when they visit a long home page or a long navigation screen. This change in behavior is probably due to users getting more experience with scrolling Web pages.
There are still a few users who rarely scroll. Even those users who are willing to scroll may be tempted to choose one of the initially visible options when it seems to match their goals. Such users will never see an even better, but invisible, choice that would have required scrolling. Therefore, I still recommend trying to design navigation pages to make all major choices visible without scrolling on the monitors used by the average visitor to a site. Also, the likelihood of making the best choice from a navigation page is maximized if the user can see and compare all the options at the same time without having to scroll and remember the hidden choices.
The change from 1994 is that scrolling is no longer a usability disaster for navigation pages. Scrolling still reduces usability, but all design involves trade-offs, and the argument against scrolling is no longer as strong as it used to be. Thus, pages that can be markedly improved with a scrolling design may be made as long as necessary, though it should be a rare exception to go beyond three screenfulls on an average monitor.
(Update 2010: Eyetracking shows that people allocate much less attention to information below the fold.)
Imagemaps Less of a Problem
Imagemaps caused endless usability problems in 1994, with users overlooking clickable areas and expressing frustration that they didn't know what they could do on the page. Imagemaps have caused very few problems in recent studies, for several reasons:
Users have gotten used to clicking on pictures that look different from standard GUI widgets
Graphic designers have gotten better at visualizing "clickability"
It is now rare for pages to consist of one huge imagemap; instead, buttons and clickable areas are more clearly delineated through the combination of multiple graphics
Client-side imagemaps allow some amount of feedback as the user moves the mouse pointer over the image. Even better feedback is needed, though.
Users have always been intolerant of downtime and crashes. As one of my very first Web test users said, "if a site doesn't work, I may give it a second chance; if it still doesn't work, I am never going back." Users also felt that "under construction" signs were disrespectful of their time.
Even though users wanted sites to work, they were willing to accept sites that had very limited functionality. In the early years, the Web was more of an experimental environment and users understood that they would be limited to a sample of a company's services. Now, users expect comprehensive service from sites. Not only do sites need to be up and available at all times, they also need to have all the information and services users want in a certain category. For example, in 1994 a publishing company could get away with featuring a few recent books on its home page and not providing access to its full backlist. Now, if a company doesn't have all its products online, users will complain.
The Web is no longer an experiment: it is mainstream. You have to rely on the ability to do business on the Web, and people get very annoyed when they can't: Limited sites are seen as a sign of corporate incompetence.