This study investigated the stability of heath care costs using a methodology featuring a two-year longitudinal design, stratified random sampling, a large sample size (N = 974), claims system data and self-report survey data, and statistical testing. The typical “regression to the mean” effect was observed, as extreme cases (both high cost and low cost) moved toward the middle during the next year. Almost two-thirds of cases changed their claims cost group status from one year to the next year. The “pareto” group (top 20% of costs in the past year) was the most unstable, with less than 4% still classified at the same highest-cost level the following year. The most striking finding was that 92% of future claims costs could not be predicted, even when using past claims costs and relevant survey data on age, sex, health care visits and psycho-social concerns. Cost control implications of prevention, health promotion, and demand management services are discussed.