The differentiation of frontal lobe epilepsy (FLE) and temporal lobe epilepsy (TLE) is a clinical problem of major theoretical and practical importance. Current electroclinical classification is based on retrospective studies of highly selected patients. When applied to the presurgical evaluation of patients, it has poor specificity. The current study adopts a different and prospective approach to the analysis of ictal clinical manifestations and their value in differentiating FLE and TLE. Two hundred and fifty-two patients with partial epilepsy were selected according to criteria of focal abnormality on imaging, ictal EEG or interictal EEG or highly focal clinical pattern. A witnessed seizure description was obtained for each of their habitual seizures and the sequence of manifestations encoded and entered into a statistical cluster analysis to form a clinical classification of the 352 seizures identified, which comprised 14 clinical groups. Neuroimaging abnormalities were measured, using a template technique, and graded 0–3 according to extent of involvement of each region in the lesion, using standard anatomical divisions. A χ2 analysis of lesion location against seizure type was performed to assess the strength of association of seizure types with specific cerebral regions. The distribution of interictal EEG spikes and ictal EEG onsets were assessed qualitatively. An independent analysis was also performed, comparing clinical seizure manifestations associated with lesions restricted to either frontal or temporal lobes. Of the 14 clinical groups, four were predominantly related to temporal lobe abnormalites: fear/olfactory/gustatory; absence with no focal symptoms; experiential and visual. Within these groups, 45 out of 58 lesional cases involved the temporal lobes (P << 0.001). A minority of seizures in these groups were associated with frontal lesions and these seizures were significantly more likely to evolve to version/posturing, without an intervening absence phase, than the temporal cases (P < 0.001). Two groups were related to perirolandic abnormalities; somatosensory and Jacksonian clonic with 22 out of 24 lesional cases involving this region (P < 0.001). Two other groups were related to the frontal lobes; version/posturing and motor agitation. Early focal tonic activity or head turning were associated with lateral premotor lesions (P < 0.001) and ictal and interictal EEG showed strong frontal predominance. Seizures characterized by general motor agitation were associated with lesions of the orbitofrontal (eight out of 13 cases) and frontopolar (six out of 13 cases) cortices (P < 0.001). Location of interictal EEG spikes and ictal EEG onsets were generally consistent with lesion sites and where there were discrepancies, EEG localization tended to be more diffuse than lesion localization, rather than frankly discordant. Analysis of manifestations associated with pure frontal and pure temporal lesions supported the results of the cluster analysis and also showed a significant association of oro-alimentary automatisms with temporal lobe abnormalities. There were no consistent differences between groups with different localizations in terms of seizure frequency or other characteristics of seizure timing, although very high seizure frequencies were seen more often in association with frontal lesions. Only one combination of different seizure types in the same patient occurred with statistical significance: absence and generalized motor seizures and pseudo generalized epilepsy. The results of this study suggest that relatively few seizures can be localized reliably on clinical grounds and that even in those seizure types where there is a statistically significant association with specific cortical areas, an important minority do not share the same associations. Analysis of the seizure evolution as well as initial symptoms may be of value in localizing some cases, but even here wide variation occurs. These findings lead us to question the value of classifiying partial epilepsy using electroclinical criteria, particularly in trying to infer anatomical localization. Possible mechanisms for blurring of the distinctions of clinical seizure types are discussed.