For a start, you might want to read through the Procedural World blog, by the guy who wrote the voxel technology that Everquest is using.

Essentially, a voxel is a 3D pixel (hence the similar naming). Just as a 2D image is constructed of pixels, a 3D image can be constructed of voxels. That's pretty much it; the idea is quite simple. The hard part comes in the implementation. Adding a dimension adds an exponential increase to the complexity of your algorithms for handling the data, as well as exponentially increasing memory storage requirements. Simple voxel systems are pretty easy, but elaborate ones such as the tech behind EQ, intended to build vast worlds, are quite complex and require a strong knowledge of parallel systems design and optimization to implement.