Decades after its first description, nonaccidental brain injury, more recently renamed ‘abusive head trauma’, is still a highly controversial diagnosis.1,2 Inevitably its social and legal implications contribute to the difficulties, but the major problem is the continuing poor understanding of its pathogenesis and mechanisms. As the currently accepted diagnostic triad (acute encephalopathy, subdural haemorrhage and retinal haemorrhages) can coincide with injuries elsewhere, there is no doubt that at least some cases result from inflicted trauma.1 Still, establishing or refuting the diagnosis in those with the triad but without other injuries remains difficult. Alternative causes are rarely identified. Shaking is the most widely accepted explanation, largely based on confession data, although the veracity of this has been queried.1–3 How shaking, or even shake-impact, could cause either the cerebral or the retinal pathology is still unclear. In children with the triad, as well as subdural collections, both neuroradiology and neuropathology most often find features of an hypoxic–ischaemic cerebral insult, and less frequently focal infarction, while other typical findings associated with trauma, such as axonal injury or tears in the brain, are relatively uncommon.3–5 This has not been satisfactorily explained. Notably, other body systems do not show asphyxial features. A further important point is that the triad is often a tetrad. A rather overlooked association is a marked anaemia, either at presentation or soon after, which is also poorly understood, since the intracranial bleeding is reported to be insufficient to account for it. It is not even certain if anaemia is a predisposing factor for, or a consequence of, the syndrome. There is now yet more to explain: a significant minority of infants diagnosed with the syndrome, some with other injuries too, have recently been shown to have spinal subdural haematomas.6 As these extend a variable distance up the spinal canal from the sacral thecal cul-de-sac they are not recognized on routine neuro-imaging which excludes the thoraco-lumbar spine. The clinical implications are unclear, but there were no obvious acute clinical or radiological signs of cord injury. Their origin is also unexplained. The authors point out that there are no bridging veins in the spinal subdural space. Based on reports in adults they suggest that the spinal subdurals are extensions of their cranial counterparts, pooling in a dependent area. In understanding the pathogenesis, animal models are unsatisfactory because of the shaking mechanisms used and the poor anatomical correlation with human infants. Mechanical models have given contradictory results, but they too suffer from anatomical deficiencies as they may not allow for difference densities inside the brain, for example between grey and white matter, or the anchoring effect of the brainstem. All the models omit possible physiological factors, as the animal models are anaesthetized and the mechanical ones are static. In confession data many infants are crying, and some are also described as coughing or choking. Crying and Valsalva type manoeuvres can be associated with a rise in intracranial pressure and/or reduced venous outflow.7 In theory, prolonged crying might have other effects which could affect brain physiology, for example on the pCO2 level, but for obvious ethical reasons this has not been studied. Not surprisingly with so many ill-understood factors the medical case that a child might have been shaken can appear weak. It is obviously important that the child should be safeguarded and here the clinical diagnosis is paramount. However, if a suspected perpetrator is facing a prison sentence it is only reasonable that these doubts should be aired. In view of the often devastating effects for the child, and the enormous amount of time and resources required in assessing this syndrome, research including systematic collections of clinical, social, radiological and pathological data, and new biomechanical models, is urgently needed.