Common problems in alignment include signal asymmetry from a nominally symmetric mark and variation across the wafer, from wafer to wafer and from mask level to mask level. Signal asymmetry arises from many sources, including uneven coating of overlayers, smearing effects of chemical mechanical polishing, or aberration and uneven illumination in the optical alignment systems. All reported alignment algorithms assume symmetric alignment signals. These alignment algorithms perform poorly when the signal is asymmetric. In previous publications [X. Chen, A. A. Ghazanfarian, M. A. McCord, and R. F. W. Pease, J. Vac. Sci. Technol. B 15, 2185 (1997); A. A. Ghazanfarian, X. Chen, M. A. McCord, and R. F. W. Pease, Proc. IEEE 4, 1913 (1998)], we proposed a new alignment method that incorporates prior knowledge by building a model for the asymmetry and process variations using alignment signals and overlay data from prior wafers: In this article, we present experimental results that convincingly demonstrate the viability of the adaptive alignment method. The experiments are carried out using deliberately distorted alignment marks placed alongside undistorted marks. It is shown that even under conditions when conventional alignment systems fail altogether, the mean-plus-three-sigma value of the alignment errors is less than 50 nm when the proposed algorithm is used.