This paper describes a technique for 3-D reconstruction of large cancellous bone regions. The output is a 3-D array describing the original cancellous bone region, and the output can be used for any kind of measurement of the bone architecture. The technique was developed as a tool for researchers conducting experimental and clinical studies related to cancellous bone architecture and, ultimately, to cancellous bone quality. A set of new and unbiased methods for quantification of cancellous bone has been a stimulus for the development of the technique, as the quantification methods rely on 3-D information. The technique is based on automated serial sectioning, and all steps from specimen preparation to image segmentation are described in detail. Examples of 3-D reconstructed vertebral bodies are given. By use of the described technique, between 300 and 600 sections can be made and registered per hour, which means that an average vertebral body can be reconstructed in about 2 h. Compared to previous implementations of the general principle of serial sectioning, this is a significant improvement both in resolution and in time efficiency.