It is known that the reconstruction produced by the standard cone beam back-projection algorithm, with circular orbit, gives only an approximation to the true three dimensional x- ray attenuation map. It is generally thought that the errors are acceptable if the cone angle is not too large. Such assumptions are based, at least in part, on reconstructions of ellipsoids or spheres. However, this is not representative of most of the specimens used for microtomography which may give rise to larger errors. We have therefore designed a mathematical cone beam phantom generator, capable of calculating a projection data set from analytical specimens described by surface polygons. Using this phantom generator, we have tested a variety of different types of 'specimen' and have shown that when they have certain characteristics, serious errors can occur with very small cone angles, while others will tolerate much larger angles. For a more authentic simulation, a polygon surface was generated from a microtomography scan of a real piece of walrus tusk. This has been used to determine the susceptibility of this type of specimen to cone-beam related errors for various cone angles.