The sources of error in cephalometric measurement and their analyses are discussed. The importance of distinguishing bias and random errors is emphasized, and methods of control are discussed. Randomization of record measurement is one of the most important methods of avoiding bias, but it is rarely undertaken in cephalometric studies. Random errors are particularly important in the evaluation of individual radiographs, and a measurement that has a high error in relation to its total variability will be of little value in clinical assessment. In serial studies of facial change, the error variance is always a major part of the total variance and thus results have to be interpreted with caution. In cross-sectional studies it is not possible to specify exactly the acceptable limits of random errors, because this will depend on the difference between groups that would be of interest and on the number of cases. The judicious replication of measurements can be important in the control of random errors. In many papers, adequate error evaluation and control is lacking. In these circumstances, the results are of limited value because it is not possible to tell whether an apparent effect is the result of bias in measurement or whether a real effect is being obscured by random errors. It is incumbent on authors to consider how their measurement errors should affect the interpretation of results.