To assess how accurate the interpretation of seizure semiology is when inferred from witnessed seizure descriptions and from video recordings, five epileptologists analyzed 41 seizures from 30 consecutive patients who had clinical episodes in the epilepsy monitoring unit. For each clinical episode, the consensus conclusions (at least 3 identical choices) based on the descriptions and, separately, of the video recordings were compared with the clinical conclusions at the end of the diagnostic work-up, including data from the video-EEG recordings (reference standard). Consensus conclusion was reached in significantly more cases based on the interpretation of video recordings (88%) than on the descriptions (66%), and the overall accuracy was higher for the video recordings (85%) than for the descriptions (54%). When consensus was reached, the concordance with the reference standard was substantial for the descriptions (k=0.67) and almost perfect for the video recordings (k=0.95). Video recordings significantly increase the accuracy of seizure interpretation.