Analyses of (palaeo)geomagnetic data can provide surprisingly strong constraints upon models of the geodynamo. In particular, it is shown here from an analysis of the geomagnetic reversal sequence that long-term variation in reversal rate does not occur as a result of change in any inherent inhibition or encouragement in the reversal process. Instead, there must be some physical process that actually triggers reversals, and this process must derive from an energy source effectively independent of the energy source powering the main geomagnetic field. From this, two very simple generalised models, referred to as the 'hot-blob' and 'cold-blob' models, are suggested. Data from individual reversal transitions are used to discriminate between these two models. The preferred model is that the geodynamo is driven primarily by convection related to freezing of the outer-core at the inner-core-outer-core boundary; cold blobs descending from the core-mantle interface are identified as the instabilities that trigger reversals.