Résumé : Fuel-lean premixed hydrogen flames are thermodiffusively unstable and cellular structures over a wide range of length-scales are formed due to strong differential diffusion of hydrogen. Pressure promotes the thermodiffusive instabilities due to the increase in Zeldovich number and the relevant reaction pathways. As the local stoichiometry and temperature correlate with the thermodiffusive instabilities, the NOx formation mechanism is expected to be complicated in fuel-lean hydrogen flames at elevated-pressure conditions. In this work, the characteristics and the NOx formation mechanism of thermodiffusively unstable premixed hydrogen flames at elevated-pressure conditions are investigated through direct numerical simulation (DNS). The designed operating conditions were motivated by findings in experiments investigating NOx emissions in pure hydrogen combustion at elevated-pressure conditions. The flame characteristics are assessed by means of the global burning velocity, flame surface area, stretch factor, and PDFs of curvature and length-scale of cellular structures. The conditioned thermo-chemical quantities are analyzed with respect to the progress variable and curvature to illustrate the key correlations. The contributions of elementary reactions to the NOx formation are quantified and the important reaction pathways are identified in a reaction pathway analysis. The global burning velocity is significantly increased with increasing pressure, which mainly results from the increased reactivity. The positive curvature is more prominent in the elevated-pressure case, which results in a higher probability of smaller cells. Different from the atmospheric case, where the NNH reaction pathway is dominant, the N2O reaction pathway is the most important under elevated-pressure conditions, which suggests the strong correlations of NOx formation with curvature values. To incorporate the wide range of curvatures in a flamelet model, an extended flamelet tabulation method based on the composition space model is proposed. The performance of the composition space model in predicting the important radicals and NOx species is evaluated through an a priori analysis. The composition space model gives better predictions of the radicals and the production and consumption rates of the NOx species compared to conventional flamelet models that do not consider the effects of local flame curvature.