The sea ice modeling community is progressing towards pan-Arctic simulations that explicitly resolve leads in the simulated sea ice cover. Evaluating these simulations against observations poses new challenges. A new feature-based evaluation of simulated deformation fields is introduced, and the results are compared to a scaling analysis of sea ice deformation. Leads and pressure ridges – here combined into linear kinematic features (LKFs) – are detected and tracked automatically from deformation and drift data. LKFs in two pan-Arctic sea ice simulations with a horizontal grid spacing of 2 km are compared with an LKF dataset derived from the RADARSAT Geophysical Processor System (RGPS). One simulation uses a five-class ice thickness distribution (ITD). The simulated sea ice deformation follows a multi-fractal spatial and temporal scaling, as observed from RGPS. The heavy-tailed distribution of LKF lengths and the scale invariance of LKF curvature, which points to the self-similar nature of sea ice deformation fields, are reproduced by the model. Interannual and seasonal variations in the number of LKFs, LKF densities, and LKF orientations in the ITD simulation are found to be consistent with RGPS observations. The lifetimes and growth rates follow a distribution with an exponential tail. The model overestimates the intersection angle of LKFs, which is attributed to the model's viscous-plastic rheology with an elliptical yield curve. In conclusion, the new feature-based analysis of LKF statistics is found to be useful for a comprehensive evaluation of simulated deformation features, which is required before the simulated features can be used with confidence in the context of climate studies. As such, it complements the commonly used scaling analysis and provides new useful information for comparing deformation statistics. The ITD simulation is shown to reproduce LKFs sufficiently well for it to be used for studying the effect of directly resolved leads in climate simulations. The feature-based analysis of LKFs also identifies specific model deficits that may be addressed by specific parameterizations, for example, a damage parameter, a grounding scheme, and a Mohr–Coulombic yield curve.