A pattern recognition algorithm is developed to estimate the top width and bottom width of trenches of semiconductor wafers simultaneously. A confocal microscope is used in this study because of its excellent depth resolution. The signals from the confocal microscope are compared to a set ofaprioriwaveforms with known trench dimensions. An optimal estimate is chosen. Principle component decomposition of the set of waveforms is used to achieve reduced storage and improved computational speed. An interpolation method is used to improve the accuracy of the estimate. Experimental results of deep trenches are provided to demonstrate the approach.