Several aspects of observed and estimated velocities for Arctic sea ice are explored. The long-term mean ice motion field and associated variance are derived using 15 years (over 85,000 observations) of daily buoy motions. The mean field is separated into the part due to long-term currents and internal ice stress and the part due to mean winds. The velocity variance is partitioned into linear and nonlinear parts. The linear response has no trend over the 15 years examined. For a 5-year subset of the buoy data, an ice-ocean model with constant air-drag parameters is used to estimate daily ice velocities at the buoy locations, and then the linear relationship between the modeled velocities and the geostrophic winds is examined. Compared with buoy velocities, the modeled velocities have a larger variance and larger linear response to the winds. In a parameter study, the effects of seasonally varying air-drag coefficients, increased ice strength, and different ocean currents are examined. Seasonally varying air drag, with smaller average magnitude than the standard value, significantly improves the match of total, linear, and nonlinear variances between observed and modeled velocities. The model simulations with varying air drag also have a smaller root-mean-square daily velocity error. Comparison of daily large-scale deformations shows that the day-to-day deformations are not modeled well but generally have means and standard deviations within ¿50% of the observed values. No one version of the model is consistently better than the others at matching the means and standard deviations of the observed daily divergence, vorticity, and shear. Ice thickness is sensitive to the model parameters varied, with 5-year average thicknesses ranging from 2.2 to 3.4 m for the various simulations. ¿ 1999 American Geophysical Union