A detailed analysis of the predictability of observed Monin–Obukhov (MO) similarity within the near-ground region of near-neutral to moderately convective atmospheric boundary layers (ABL) from large-eddy simulation (LES) fields is reported in this work. High-resolution LES predictions of means, variances, budgets of turbulent kinetic energy and temperature variance, and the velocity and temperature spectra from three ABL states (−zi/L=0.44, 3 and 8) are analysed under MO scaling. The resolution in the near-ground region is increased by using ‘nested meshes.’ For the close-to-neutral case (−zi/L=0.44) the relative roles of grid resolution and subgrid-scale scale (SGS) parameterization on the predictability of MO-similarity are also studied. The simulated temperature field is found to satisfy the MO hypothesis and agree well with observations. The simulated velocity field, on the other hand, shows significant departures. Except for the horizontal variance, MO scales are the appropriate normalizing scales for the near-ground-layer statistics. However, the LES suggest that the boundary layer depth zi has an ‘indirect’ influence on all near-ground-layer variables except temperature, and the LES-predicted MO-scaled variables exhibit a functional dependence on both z/L and z/zi. The simulated two-dimensional spectra of velocity and temperature fluctuations, however, suggest that while large scales deviate from MO-similarity, inertial subrange scales are MO-similar. Discrepancies with field observations raise important questions of the non-dimensional depth z/zi over which MO-similarity holds for a particular variable. Surface-layer field studies generally do not document zi. It is also not clear to what extent these discrepancies are due to approximations made in LES. Measurements are needed designed specifically for comparing with LES predictions.