A computer-implemented method of providing and selecting multiple text prediction from text entered from multiple windows of a computer system using an application-independent text prediction system. Application-independence is the ability to work with many different applications without being customized for each applications. Because different prediction methods are appropriate for different applications, multiple prediction components called "prediction modules" provide text predictions based on each prediction module's prediction method from which the best prediction may be displayed. The best prediction from the results of the multiple methods may be displayed. In accordance with providing multiple predictions from various modules, a method is provided for choosing which modules' predictions should be used from the set of possibly contradictory predictions. Each prediction module produces a weighted list of predictions. The prediction module weights are based on the prediction modules' estimation of the probability that the completion text will be entered by the user. The resulting best prediction or predictions with the greatest weights may then be presented to the user.