Context. During the last decade, the FORS1 instrument of the ESO Very Large Telescope has been extensively used to study stellar magnetism. A number of interesting discoveries of magnetic fields in several classes of stars have been announced, many of which obtained at a ∼3σ level; some of the discoveries are confirmed by measurements obtained with other instruments, some are not. Aims. We investigate the reasons for the discrepancies between the results obtained with FORS1 and those obtained with other instruments. Methods. Using the ESO FORS pipeline, we have developed a semi-automatic procedure for magnetic field determination. We have applied this procedure to the full content of circular spectropolarimetric measurements of the FORS1 archive (except for most of the observations obtained in multi-object spectropolarimetric mode). We have devised and applied a number of consistency checks to our field determinations, and we have compared our results to those previously published in the literature. Results. We find that for high signal-to-noise ratio measurements, photon noise does not account for the full error bars. We discuss how field measurements depend on the specific algorithm adopted for data reduction, and we show that very small instrument flexures, negligible in most of the instrument applications, may be responsible for some spurious field detections in the null profiles. Finally, we find that we are unable to reproduce some results previously published in the literature. Consequently, we do not confirm some important discoveries of magnetic fields obtained with FORS1 and reported in previous publications. Conclusions. Our revised field measurements show that there is no contradiction between the results obtained with the low-resolution spectropolarimeter FORS1 and those obtained with high-resolution spectropolarimeters. FORS1 is an instrument capable of performing reliable magnetic field measurements, provided that the various sources of uncertainties are properly taken into account.