Abstract In Part I of this paper, we ,summarize the principal results of our model atmosphere computations for early-type stars that cover the main-sequence A0~05 stars. These are constructed with the use of the Rosseland mean. There are much differences in structure of the model atmospheres that are constructed with the use of other definitions of opacity. Some characteristics for each model atmosphere are pointed out and discussed. Part II contains some discussions in connection with the choice of the opacity. We shall compare the three model atmospheres which are constructed with the similar conditions of Te = 20,400° K and log g = 3.8, but with the use of different definitions of opacity, i.e., Rosseland, Chandrasekhar, and Planck mean, respectively. Some remarks on the structure of ea.ch model atmosphere are. shown and discussed. Moreover we discussed the definitions of the opacity and the revision for them. We defined the revised expression of the opacity κ−* with the formula, κ−*=∫κνFνdν/∫Fνdν. This revised value κ−* of opacity should be close to the assumed one, if adopted model atmosphere is a better representation. For the model atmosphere with the Rosseland mean, the ratio κ−R*/κ−R of the revised value κ−R* to the adopted Rosseland mean κ−R is closer to unity than any other definitions of opacity. As the concluding remarks, we discussed the advantage of the Rosseland mean in constructing early-type model stellar atmospheres.