This study surveys and evaluates similarity theory for estimating the sea-surface drag coefficient with the bulk aerodynamic method. The most commonly used formulations of the aerodynamic roughness length, required by similarity theory, are examined using data sets from four different field programs. These relationships include the Charnock formulation and the wave age modified Charnock relationship. The goal is to assess the overall performance of simple formulations of the roughness length including cases where the Charnock formulation is not expected to apply, and to assess the errors resulting from application of the Charnock formulation to all conditions, as is done in many numerical models where an explicit wave model cannot be accommodated. This examination indicates that spurious self-correlation explains more variance than actual physical relationships, even after eliminating weak wind cases. Frequent cases of anomalously low stress and very small values of the Charnock coefficient further reduce the usefulness of this formulation for the present data sets. Causes of the frequent very small values of the Charnock coefficient are briefly investigated.