Real time cephalometry has been compared with B scan cephalometry by a rigorous method with complete partition of variables. Analysis was by standard analysis of variance techniques. Systematic and random errors between and within observers, machines and days have been evaluated. The largest systematic error has been shown to be between machines. This is due to failure to align the beam precisely at right angles to the mid-line of the fetal head using B scanners. Real time scanners with bar calipers are, therefore, superior to B scanners for cephalometry. There is a lesser difference between observers. An additional element of variability arises when measurements are made on different days but this is smaller than other sources of variation and smaller than the random variation. Random errors are described by a standard deviation of 1 mm and systematic errors are all numerically smaller than this.