In order to obtain the highest possible accuracy from GPS measurements, in particular needed for deformation analysis, carrier phase observations are used. Within the mode of double difference carrier phase, the fundamental mixed integer-real valued adjustment problem is met. Through standard parameter estimation like weighted least squares or alternative robust objective functions only the »floating solution« – including the vector of ambiguity – can be given. In order to speed up the searching process for the integer values of ambiguity, the method of decorrelation is applied which works »in practice« sufficiently well. Here we evaluate three proposals for the decorrelation of float solutions for ambiguity resolution, namely (i) the inverse integer Cholesky decorrelation (CHOL) as proposed by P. Xu (2001) (ii) the integer Gauss decorrelation (GAUSS) initiated by P. Teunissen (1997) and (iii) the A. K. Lenstra, H. W. Lenstra, and L. Lovacs (LLL) algorithm as proposed by A. Hassibi, S. Boyd (1998), and E. Grafarend (2000). The analysis of different decorrelation methods is made as realistic and as statistically meaningful as possible: a random simulation approach has been implemented which guarantees a symmetric, positive definite variance covariance matrix of the ambiguity vector (»float solution«) derived from the double difference observation variance covariance matrix. The spectral condition number is used as a criterion for the performance of the three decorrelation methods. Three sets of simulated data are finally comparatively evaluated.