Test-retest reliability of rotational chair testing for a single facility has previously been examined by others. The actual data analysis methods, however, have received far less attention. The variety of both hardware and software currently used theoretically may affect the results for a given subject tested at different facilities. The purposes of this study were, first, to quantify the amount of variability in the analysis of identical raw data files at multiple rotational chair testing facilities by using automated analysis; second, to evaluate the effect of operator intervention on the analysis; and third, to identify possible sources of variability. Raw data were collected from 10 normal subjects at 0.05 Hz and 0.5 Hz (50 degrees per second peak velocity). Diskettes containing raw electro-oculogram data files were then distributed to eight participating laboratories for analysis by two methods: (1) using automated analysis algorithms and (2) using the same algorithms but allowing operator intervention into the analysis. Response parameters calculated were gain and phase (re: velocity). The SD of gain values per subject for automated analysis ranged from 0.01 to 0.32 gain units and of phase values from 0.4 to 13.7 degrees. For analysis with operator intervention, the SD of gain values ranged from 0.02 to 0.10 gain units and of phase values from 0.4 to 4.4 degrees. The difference between automated analysis and analysis with operator intervention was significant for gain calculations (p < 0.02) but not for phase calculations (p > 0.05). This study demonstrates significant variability in automated analysis of rotational chair raw data for gain and phase.(ABSTRACT TRUNCATED AT 250 WORDS)