A clinician should not rely entirely upon a caregiver's report regarding behavioral pathology when planning a treatment strategy. Direct observational evaluation instruments as well as caregiver-based assessments are necessary. A new scale for the empirical (observational) evaluation of behavioral symptoms in Alzheimer's disease (AD) and related dementias, the Empirical Behavioral Pathology in Alzheimer's Disease Rating Scale (E-BEHAVEAD) was developed. Interrater reliability of this new assessment instrument was examined. Additionally, the relationship between the observed occurrence of behavioral symptomatology on this new rating instrument was compared with the occurrence using a similarly designed, caregiver-based instrument. The interrater reliability study consisted of two raters who simultaneously evaluated 20 dementia patients. The comparative study employed a cross-sectional design (N = 49). Individuals were evaluated in an outpatient clinic setting. The study population consisted of cognitively normal individuals and dementia patients. Evaluations included the new, observationally based behavioral assessment (the E-BEHAVE-AD), a caregiver-based behavioral assessment (the Behavioral Pathology in Alzheimer's Disease Rating Scale; BEHAVE-AD), a clinical global measure (the Global Deterioration Scale), and a mental status assessment (the Mini-Mental State Examination). The interrater reliability study revealed an intraclass correlation coefficient of .97 (p < .01) for total scores on the new E-BEHAVE-AD rating scale. The correlation coefficient for the amount of agreement on the presence of symptoms in six symptomatic categories between caregiver-based information about the patient's behavioral pathology assessed on the BEHAVE-AD and the clinician's observations assessed with the new E-BEHAVE-AD rating instrument was .51 (p < .01). The New E-BEHAVE-AD rating instrument showed excellent interrater reliability. Furthermore, there was a statistically significant relationship between clinician observation of the occurrence of behavioral pathology assessed using the E-BEHAVE-AD and caregive-reported pathology assessed with the BEHAVE-AD. However the magnitude of the correlation between these measures indicated that the majority of variance was independent and nonoverlapping. Consequently, these data support theoretical models suggesting that the assessment of behavioral pathology in dementia might ideally encompass both direct observational and caregiver-report approaches, using measures such as the E-BEHAVE-AD as well as measures such as the BEHAVE-AD.