For the last twenty years, human-computer interfaces have been dominated by two-dimensional interaction techniques. Things are changing. Techniques that were previously restricted to specialised CAD/CAM tools and immersive VR systems are now being extended to the mass market. The photo-realistic facilities offered by QuicktimeVR and the model based renderings of VRML (Virtual Reality Mark-up Language) provide sophisticated tools for interface design. As a result, three dimensional visualisation techniques are being widely exploited in the financial services industry, airports and even off-shore oil production. Unfortunately, research in human- computer interaction lags well behind commercial practice. There are few guidelines that can be applied to support the development of these 3D interfaces. In consequence, users often report intense frustration as they navigate around virtual information spaces. This paper, therefore, describes a number of evaluations that have been conducted to examine the usability problems that affect these interfaces. It is concluded that the standard measures of task performance and subjective satisfaction cannot easily be applied to assess the utility of 3D systems. Finally, Gibson's work on direct perception is used to explain why people find it difficult to identify the underlying usability problems that affect this new generation of human-computer interfaces.