Ictal patterns on scalp-electroencephalography are often visible only after propagation, therefore rendering localization of the seizure onset zone challenging. We hypothesized that mapping haemodynamic changes before and during seizures using simultaneous video-electroencephalography and functional imaging will improve the localization of the seizure onset zone. Fifty-five patients with ≥2 refractory focal seizures/day, and who had undergone long-term video-electroencephalography monitoring were included in the study. 'Preictal' (30 s immediately preceding the electrographic seizure onset) and ictal phases, 'ictal-onset'; 'ictalestablished' and 'late ictal', were defined based on the evolution of the electrographic pattern and clinical semiology. The functional imaging data were analysed using statistical parametric mapping to map ictal phase-related haemodynamic changes consistent across seizures. The resulting haemodynamic maps were overlaid on co-registered anatomical scans, and the spatial concordance with the presumed and invasively defined seizure onset zone was determined. Twenty patients had typical seizures during functional imaging. Seizures were identified on video-electroencephalography in 15 of 20, on electroencephalography alone in two and on video alone in three patients. All patients showed significant ictal-related haemodynamic changes. In the six cases that underwent invasive evaluation, the ictal-onset phase-related maps had a degree of concordance with the presumed seizure onset zone for all patients. The most statistically significant haemodynamic cluster within the presumed seizure onset zone was between 1.1 and 3.5 cm from the invasively defined seizure onset zone, which was resected in two of three patients undergoing surgery (Class I post-surgical outcome) and was not resected in one patient (Class III post-surgical outcome). In the remaining 14 cases, the ictal-onset phase-related maps had a degree of concordance with the presumed seizure onset zone in six of eight patients with structural-lesions and five of six non-lesional patients. The most statistically significant haemodynamic cluster was localizable at sub-lobar level within the presumed seizure onset zone in six patients. The degree of concordance of haemodynamic maps was significantly better (P < 0.05) for the ictal-onset phase [entirely concordant/concordant plus (13/20; 65%) + some concordance (4/20; 20%) = 17/20; 85%] than ictal-established [entirely concordant/concordant plus (5/13; 38%) + some concordance (4/13; 31%) = 9/13; 69%] and late ictal [concordant plus (1/9; 11%) + some concordance (4/9; 44%) = 5/9; 55%] phases. Ictal propagation-related haemodynamic changes were also seen in symptomatogenic areas (9/20; 45%) and the default mode network (13/20; 65%). A common pattern of preictal changes was seen in 15 patients, starting between 98 and 14 s before electrographic seizure onset, and the maps had a degree of concordance with the presumed seizure onset zone in 10 patients. In conclusion, preictal and ictal haemodynamic changes in refractory focal seizures can non-invasively localize seizure onset at sub-lobar/gyral level when ictal scalp-electroencephalography is not helpful.