MR data, direct from the scanners, are usually stored as two byte short integers, as this reflects the resolution of today’s analog to digital convertors, and allows a signal to noise ratio of about 60,000:1; substantially better than the native MR signal. Typical functional data sets, however, track only the temporal changes in the signal intensity, which are usually much smaller than the base signal. In most cases, in our experience, the magnitude of the signal change can be described fully in one byte of the two byte data. Even when the functionally-significant changes are large, most of the pixel locations are either image background or changing very little. Were the actual data scaled such that the full range of the short integers was used, this would only allows us to track changes of about 0.4% (1 part in 256). In practice, however, the images are usually scaled to use only a fraction of the available dynamic range, using mean values of about 1000 or 2000 (arbitrary signal units) to accommodate the largest possible signals without digital saturation or wrap.