Interobserver reliability in obtaining neurologic histories and examinations was investigated among neurologists collaborating in the Stroke Data Bank (SDB). Seventeen in-hospital stroke patients were examined by six neurologists experienced in stroke over the course of three days. Patients were examined twice a day for two successive days, with each patient seen by four different neurologists. Data were recorded on SDB forms, according to definitions and procedures established for the SDB. Percent agreement and kappa coefficients were calculated to assess the levels of agreement for each item. Important differences in levels of agreement were found among items on both neurologic history and examination. Agreement among neurologists was higher for neurologic examination than for history. Patterns of agreement for items with low prevalence or with numerous unknown ratings are discussed. Improvement in interobserver agreement due to data editing for intra-observer consistency was shown.