For people with motor disabilities unable to control a pointing device, there is a need for an efficient keyboard Web navigation method. The current state of keyboard navigation tools is surveyed and discussed. The standard tab key navigation system is widely regarded as insufficient for practical keyboard access to the Web. Other techniques, such as identifier navigation --- where links are activated by entering their numeric code --- are found to be much more efficient, but possibilities remain for improvement. A technique based on selecting links by text search provides an alternative method of keyboard navigation, but does not minimize the number of required keyboard inputs, which is important for individuals not capable of fast typing input. A new technique called KeySurf is proposed, which aims to make text search navigation more efficient and intuitive by estimating which elements are more likely to be selected by the user, and then allowing those elements to be selected with fewer keystrokes. Web page elements without text associated with them are assigned labels based on algorithms that make generated labels predictable to the user. Various constraints on the searching algorithm are implemented to decrease the keystroke cost of selections, such as selecting visible elements first, matching the first characters of labels, and prioritizing visually prominent elements. In addition, the user's browsing history is used to calculate a measure of page and element interest in order to make interesting elements easier to select. KeySurf performance is examined in three experiments: an automated analysis of keystroke cost of element selection on randomly selected Web pages, a small study involving four individuals with motor disabilities to compare KeySurf and mouse use, and a study of 11 individuals browsing the Web with KeySurf collecting data passively in the background. The automated selection test calculates the number of keys necessary to activate each link for 48,182 links, resulting in a mean of 2.69 keystrokes. The study involving individuals with disabilities shows that KeySurf can be faster than mouse use if the user is able to type 2 or 3 keystrokes faster than pointing to a target using their pointing device. The study with 11 non-disabled individuals shows that for 4,601 recorded clicks, KeySurf would have required 2.38 keys per selection. Comparing mean keystroke cost for pages containing similar numbers of elements, we find that for real Web sessions KeySurf can decrease keystroke cost by 15\% compared to the simulation results by anticipating which elements a user is likely to select. A keystroke level model of tabbing, ID navigation, and KeySurf indicates that the predictability of ID navigation makes it more efficient for faster typists, while KeySurf is likely to be faster for slower typists. / Graduate