The purpose of this thesis was to find and implement an adaptive method, based on given data and hardware, for selecting different level-of-detail whilst preserving visual quality to the best extent possible. Another important aspect of the new method was that it had to be performance effective, since the target platform was an interactive direct volume rendering application. The project was targeted towards memory restricted systems in which it has previously been impossible to render large-scale volumetric datasets. The previous state of the target platform supports two detail levels: full and half, and does not implement any kind of prioritisation when selecting the level-of-detail of bricks. Apart from failing to render parts of the datasets, the old implementation is also lacking in that the top of the dataset always has the lowest prioritisation, which can prove problematic for certain datasets. An adaptive method which determines a suitable number of detail levels at run-time has been implemented. The new implementation has also reworked the way bricks are prioritised during rendering. The proposed algorithm prioritises bricks holding surface information as well as bricks that match the transfer-function configuration well. The results show that the proposed method is able to render large-scale datasets in limited environments whilst maintaining interactive frame-rates. The new brick selection algorithm is a step in the right direction towards solving the issue with parts of the dataset not being prioritised.