For more than a decade now, computer simulations of sound fields in rooms have been widely adopted in research and for consulting purposes. Most computer simulations are either based on geometrical room acoustics or statistical methods, hereby neglecting diffraction and interference effects. The calculation algorithms in this type of simulations often combine the image-source method and the ray-tracing technique. In this paper, the acoustics of an open-air roman theatre are investigated. This is a special case which sets up a challenge to these prediction methods. The absence of a roof and therefore of a reverberant field, demands high accuracy in predicting the early reflections. The energy dissipates quickly in this type of enclosures and there is little masking effect of the reverberation. The inverse cone shape of these theatres also puts serious limitations to the image-source method, where great areas are in the shadow zone of the mirroring surface. Another aspect that has been shown to give very different results in this case study, is the geometrical detailing of the models. Although it has been pointed out in several studies that models with a limited number of surfaces give more accurate results, in this particular case the opposite is shown to be true. The aim of this paper is to clarify some of the problems that can arise in this type of constructions, and give guidelines for how they can be overcome / avoided. Another objective is to emphasize that room acoustic computer simulations although very useful need careful consideration about the underlying calculation methods.