A comparative analysis of three recent and reliable SnIa datasets available in the literature was made: the Full Gold (FG) dataset (157 data points 0 < z < 1.75), a Truncated Gold (TG) dataset (140 data points 0 < z < 1) and the most recent Supernova Legacy Survey (SNLS) dataset (115 data points 0 < z < 1). It was found that the best fit dynamical w(z) obtained from the SNLS dataset does not cross the PDL (w −1) and remains above and close to the w −1 line for the whole redshift range 0 < z < 1. In contrast, the best fit dynamical w(z) obtained from the Gold datasets (FG and TG) clearly crosses the PDL and departs significantly from the w −1 line, while the LCDM parameter values are about 2 σ away from the best fit w(z), however not excluding it. Also for the Gold dataset a scalar-tensor theory was tested, for which the best fit form of w(z) was also found to cross the phantom divide line.