The improvement of input efficiency of graphical user interfaces (GUIs) is an important issue in human computer interaction research. The existing researches include two aspects: pointing techniques and adaptive user interfaces: the formal manipulates either the visual representation or the controlling method of a cursor while the latter adjusts the layout of widgets. However, both approaches suffer drawbacks. This paper analyzes the operations on GUI and presents a quantitative model to evaluate the input efficiency of GUIs. Based on the metric, the study then proposes a novel approach, the adaptive cursor, which adaptively supports a predicted set of widgets with pointing techniques to enable fast access. This approach avoids the extra cognitive cost caused by adaptive user interfaces, which frequently adjust the widget layout. It also extends previous pointing techniques that should have applied only to a sparse layout. To evaluate its usability, the study realizes the adaptive cursor technique on Visual Studio, whose interface contains a considerate number of controls. Experimental result shows that the adaptive cursor can save up to 27.7% of pointing time.