Variance and covariance estimation (VCE) has been attracting a lot of interesting in geodesy or other branches of learning. There are some kinds of methods that have been put forward to conduct estimation of variance and covariance components in Gauss-Markov model (GM). The VCE methods now are for models in which coefficient matrix A is free of contamination, and different with GM model, in errors-in-variables model (EIV) the design matrix A is also influenced by random noises EA, i.e. y=(A-EA)x+ey. If errors of observations in y and A are not independent and maybe correlated in some degree in some cases, the correlation matrix QyA is of real sense, i.e. QyA????0. Whether the variance and covariance components (Qyy, QAA, QyA) in EIV model can be estimated and the VEC method now can be used to estimate the components or not? In this paper we discuss the VCE problem in EIV model on the basis of the linearized EIV model and we find that the variance and covariance in EIV model cannot be estimated and only an integrated term by them can be got. The reason for the inestimability is that information in observation vector y and matrix A cannot be distinguished from each other with the linearized model and no adequate information can be used to carry out the VCE in EIV model.

Keywords: variance and covariance, variance and covariance estimation, errors-in-variables model