Atomic data in non-LTE model stellar atmospheres

We review the sources and the handling of atomic data in NLTE line-blanketed model stellar atmospheres, with an emphasis on our code TLUSTY. Incorporating tens of thousands of energy levels and millions of lines explicitly in NLTE is now feasible, but requires a statistical approach. We discuss our implementation of superlevels, the representation,of line opacity by Opacity Distribution Functions and by Opacity Sampling, and the concept of Resonance-Averaged Photoionization cross-sections. We show that different implementations of bound-bound collisional strengths in Fe superlines may result in systematic effects in the predicted emergent spectrum. Finally, we present a short study of the influence of various assumptions in the Fe model data on the resulting model atmospheres.