An algorithm for the identification of areas that do not necessarily represent the true sample surface in atomic force microscope images is presented. These areas describe regions of the surface which might not have made contact with the probe tip during a raster scan, giving data which should be deemed questionable. Through the identification of these questionable data points, a more accurate picture of the sample can be obtained. The procedure is applied to several atomic force microscope images for the improvement of sample images and for obtaining tip information. While the algorithm is applicable to all such images, its sensitivity to noise reflects shortcomings in the assumption made in deconvolution.