Representations of the reversing geomagnetic field are generally based on paleomagnetic directions or virtual geomagnetic poles, i.e., one coordinate related to field intensity is missing. Interpretations based on these representations have led to conflicting models of reversals. We have assembled records of reversals for which an intensity determination is available, either absolute (volcanic sequences) or relative (sedimentary sequences), and we have chosen to represent them in terms of directions in local Cartesian coordinates: this representation is identical to the Zijderveld representation which is traditional in paleomagnetism. The 38 selected records correspond to Matuyama-Brunhes, Jaramillo, Olduvai and some older reversals up to 5 m.y. old. A total of 22% are from volcanic sequences and the rest from sedimentary ones. Vector behavior falls into one of five categories: A, the reversing field follows the local trend of the geocentric axial dipole (GAD) passing through the origin, with noise added; B, same as A, but offset by a constant component from the origin; C, the reversing field follows a linear trend going through the origin in Cartesian coordinates but the direction is different from that of GAD; D, same as C but offset from the origin by a constant component; E, large noise, no linear or other simple behavior. Over half (55%) of the records fall in either category A or B, 18% in category E. The C-type data often consist of dipole-parallel segments separated by jumps, the significant of which is as yet unclear. The number of cases with a linear behavior is remarkable (82%). Large directional swings in the traditional representations often correspond to far less conspicuous changes on the vector plots at points close to the origin; behavior there is dominated by ''noise'', one source of which may simply be secular variation of the nondipole field. The constant offsets sometimes observed (34% of the cases) are as yet unexplained but may be due to rock magnetic problems and therefore may have no global significance. We doubt whether the present data allow us to extract resolvable terms such as g02 or h13 which have been inferred from the directional representation. ¿ American Geophysical Union 1993