Although studies of occupational exposure to volatile organic compounds (VOCs) often partition variability across groups, and between and within persons, those of environmental exposure to VOCs have not involved such partitioning. Using data from the Environmental Protection Agency's total exposure assessment methodology (TEAM) studies, we partitioned exposure variability across cities, and between and within persons for nine VOCs. The estimated variance components decreased in the order: within-person > between-person > across city. Despite their smaller magnitudes, estimates of between-person and across-city variance components were sufficiently large to provide reasonable contrast for informative epidemiology studies of most VOCs. Estimates of between-person variance components for environmental VOCs were similar to those published for occupational VOCs (groups defined by job and factory). However, estimates of within-person variance components were much greater for environmental VOCs, probably due to the greater diversity of locations (including the workplace) visited by the general public over time. For benzene and perchloroethylene, we used a simple model to calculate numbers of personal measurements required to relate the exposure level to health outcome statistically. About 10 times more personal measurements would be required to investigate perchloroethylene exposure as compared to benzene exposure; this disparity reflects the greater within-subject variability of perchloroethylene data compared to benzene data. We conclude that variability should be partitioned for environmental VOC exposures in much the same manner as for occupational exposures. There should be sufficient variability in the levels of most VOCs across cities and between subjects to provide reasonable contrast for informative epidemiology studies, as we illustrate for exposures to benzene. Yet, epidemiologists should be wary of investigating environmental VOCs without preliminary data with which to estimate the variance structure of exposure variables.