A dipole model of the P300 component of the auditory evoked potential was developed with the aim of separating the overlapping P300 subcomponents and assessing the reliability of P300 dipole parameters compared with other methods such as principal component analysis (PCA). P300 was recorded within an auditory oddball paradigm in 57 healthy subjects. Dipole source analysis was performed with Brain Electrical Source Analysis (BESA). With a temporo-basal and a temporo-superior dipole per hemisphere, most of the variance of scalp-recorded P300 activity can be explained. When dipole source analysis is used, the test-retest reliability of P300-amplitude measurements is enhanced as compared with PCA or single channel analysis. The P300 activity of the temporo-basal dipoles and that of the temporo-superior dipoles differ in several respects (e.g. latency, correlations with age) and thus appear to reflect functionally different neuronal processes. In spite of problems in the interpretation of dipole locations, it is concluded that dipole source analysis enhances both the reliability and physiological validity of P300 parameters.