The top of Earth's liquid outer core is nearly 2900 km beneath Earth's surface, so we will never be able to observe it directly. This hot, dense, molten iron-rich body is continuously in motion and is the source of Earth's magnetic field. One of the most dynamic manifestations at Earth's surface of this fluid body is, perhaps, a reversal of the geomagnetic field. Unfortunately, the most recent polarity transition occurred at about 780 ka, so we have never observed a transition directly. It seems that a polarity transition spans many human lifetimes, so no human will ever witness the phenomenon in its entirety. Thus we are left with the tantalizing prospect that paleomagnetic records of polarity transitions may betray some of the secrets of the deep Earth. Certainly, if there are systematics in the reversal process and they can be documented, then this will reveal substantial information about the nature of the lowermost mantle and of the outer core. Despite their slowness on a human timescale, polarity transitions occur almost instantaneously on a geological timescale. This rapidity, together with limitations in the paleomagnetic recording process, prohibits a comprehensive description of any reversal transition both now and into the foreseeable future, which limits the questions that may at this stage be sensibly asked. The natural model for the geomagnetic field is a set of spherical harmonic components, and we are not able to obtain a reliable model for even the first few harmonic terms during a transition. Nevertheless, it is possible, in principle, to make statements about the harmonic character of a geomagnetic polarity transition without having a rigorous spherical harmonic description of one. For example, harmonic descriptions of recent geomagnetic polarity transitions that are purely zonal can be ruled out (a zonal harmonic does not change along a line of latitude). Gleaning information about transitions has proven to be difficult, but it does seem reasonable to draw the following conclusions with varying degrees of confidence. There appears to be a substantial decrease in the mean intensity of the dipole field during a transition to ~25% of its usual value. The duration of an average geomagnetic polarity transition is not well known but probably lies between 1000 and 8000 years. Values outside these bounds have been reported, but we give reasons as to why such outliers are likely to be artifacts. The reversal process is probably longer than the manifestation of the reversal at Earth's surface as recorded in paleomagnetic directional data. Convection hiatus during a geomagnetic polarity transition seems unlikely, and free-decay models for reversals appear to be generally incompatible with the data. This implies that certain theorems in dynamo theory, such as Cowling's theorem, should not be invoked to explain the origin of reversals. Unfortunately, the detailed description of directional changes during transitions remains controversial. Contrary to common belief, certain low-degree nondipole fields can produce significant longitudinal confinement of virtual geomagnetic poles (VGP) during a transition. The data are currently inadequate to refute or verify claims of longitudinal dipole confinement, VGP clustering, or other systematics during polarity transitions. ¿ 1999 American Geophysical Union