The least squares estimator, as an optimal estimator, has been widely used in the data processing of surveying science. Unfortunately the reliability of the least squares estimator may easily be destoryed by gross errors since its mathematic model is based on the prior assumption that errors of observations must be normally distributed. To avoid this problem, the robust estimator has been created and applied to various fields. The repeated median estimator (belongs to robust estimator family) has been studied in this article. An example of the transformation of coordinates between two systems using the repeated median was chosen to illustrate its principle and applications.