The formal errors on the astrometric data in the 1997 publication of the Hipparcos catalogue are, for stars brighter than about magnitude 9, largely the result of inaccuracies in the description of the along-scan attitude of the satellite. A detailed study of the dynamics of the Hipparcos satellite has led to a much improved understanding and modelling of the satellite attitude, taking into account peculiarities in the rotation of the payload (scan-phase jumps) as well as the detections of small hits. A new reduction of the Hipparcos data was initiated, in which the attitude modelling is a direct description of the dynamics of the satellite. In this so-called fully-dynamic attitude modelling the underlying torques acting on the satellite are reconstructed, and rates and error angles are obtained through integrations. Both the hits and the scan-phase jumps could be taken into account in the context of this model. The new model, including the provisions for discontinuities, led to a factor five reduction in the attitude noise. A new reduction, based on a global iterative solution like is also planned for Gaia, was started in 2004, and completed, after some 15 iterations, in 2007. In the process of this reduction, also the sensitivity of the solution to optimal connectivity between the data in the two fields of view was exposed and taken care of, as well as a couple of small-scale calibrations. The latter had not been possible to solve for in the original reductions. In the catalogue that resulted from this new reduction, the errors on the astrometric data for all but stars brighter than magnitude 4, are dominated by photon noise. Error-correlation levels in the underlying abscissa data are down by more than an order of magnitude, and play no longer any significant role. This feature very much simplifies the analysis of, for example, wide-binary stars and open cluster data.