A series of correlation experiments and regression models are used to explore the atmospheric forcing of observed interannual variability in North Pole ice drafts as measured by U.S. Navy submarine sonar from 1982 to 1992. Mean ice drafts as well as ice draft distributions are used in the analysis. Nearby drifting buoys are used to approximate a full 2-year motion history of the North Pole ice parcels prior to submarine underpass. Drawing from theoretical work and empirical observation, predictors of North Pole ice draft variation are selected which embody accumulated ice exposure to subfreezing air temperature, ice divergence, and patterns of ice motion. The timescales of the predictors are based on systematic evaluations of the lag dependence of the predictor-predictand relationships. Linear correlation of each predictor with the observed ice draft distribution shows accumulated ice exposure to subfreezing air temperature is weakly correlated with reduction in shallow (0 to 2.5-m) drafts. Ice divergence a weak prior to draft measurement correlates most closely with the redistribution of ice drafts shallower than 3.5 m. Ice divergence and ice deflection from its wind-forced motion during the 6 months prior to observation correlate with deeper ice drafts (deeper than 6.8 and 4.6 m, respectively). Together, these two parameters explain more than 80% of the variance of the observed mean ice draft. The accumulated ice exposure to subfreezing air temperatures colder than the 1980--1992 mean in the 2 years prior to ice arrival at the North Pole adds little to an explanation of the interannual variance of the mean ice draft. The results imply that (1) a realistic formulation of sea ice dynamics is a first-order priority for model simulations of ice draft and (2) North Pole ice thickness is not a simple function of Arctic air temperature. ¿ 1999 American Geophysical Union