The results of flash visual evoked potentials (VEPs) in 44 infants blind or severely visually impaired from non-ocular causes are presented, and related to the subsequent visual outcome. Ocular causes of visual impairment were excluded by clinical examination and electroretinography. Using a 2 x 2 contingency table, a significant association between VEP and outcome was demonstrated (chi 2 = 3.51, 1 d.f., p = 0.05). Of 13 infants with normal VEPs, 11 demonstrated substantial visual improvement (negative predictive value = 84.6%). However, of the 31 with abnormal VEPs, only 14 remained severely impaired/blind; the other 17 demonstrating visual improvement (positive predictive value = 45.1%). The sensitivity of the method was high in that 14 of 16 (87.5%) infants who remained impaired/blind had abnormal VEPs, but specificity was low as only 11 of 28 (39.3%) who showed visual improvement had normal VEPs. The accuracy of the technique was therefore low, 25 of 44 (56.8%) being true positive/ negative. With regard to visual outcome when faced with an apparently blind infant, it is important not to be too pessimistic for, as is shown in this study, 28 of 44 demonstrated substantial improvement. There are no absolute indicators of prognosis, but the presence of structural cerebral lesions and a history of either neonatal meningitis or encephalopathy are relatively bad prognostic signs. The flash VEP, despite its limitations, is a useful prognostic tool, particularly in those apparently blind infants whose normal ocular examination/electroretinogram is accompanied by normal VEPs. Those with abnormal VEPs, however, do not necessarily have a poor prognosis, but should be followed-up as maturational changes and/or improvements in function of the sensory pathway will be reflected in the evoked potentials.