Abstract: The robust Least Squares (LS) adjustment has been extensively studied and successfully applied in the real applications to resist the influence of gross errors in observation. However in the LS adjustment, the coefficient matrix is considered as non-error. Unfortunately variables in coefficient matrix are inevitably error-contaminated when they come from the real observations. Considering the random errors and gross errors may exist both in the observation vector and the coefficient matrix, the robust Total Least Squares (TLS) solution is studied in this paper. The reweighting iteration method is used to detect and identify the gross errors. At the end of the paper, the simulated experiments of three-dimensional similarity coordinate transformation are investigated to prove that using the robust TLS method, the influence of the gross errors can be resisted, and the reliable solution can be obtained.