Three models of intraindividual variation are reviewed, and statistical methods for distinguishing among them are discussed. Application of these methods to short series of observations from healthy individuals indicates that, in the large majority of cases, a strictly homeostatic model is appropriate for such constituents as serum calcium and magnesium. In less closely controlled variables, e.g., serum cholesterol and uric acid, a nonstationary, "rndom walk" model appears moresuitable in most cases. A more general autoregressive model, which includes the other models as extreme cases, could be used to describe all degrees of homeostatic control. This model is more complex, however, and requires at least 10 observations to yield estimates of acceptable precision. Moreover, it is sensitive to fluctuations in within-batch analytical variance. When biological variance is small relative to analytical variance, all three models yield essentially the same predicated values. To illustrate their use, these models have been applied to four short individual series of cholesterol observations showing increasing amounts of intrapersonal variation over long periods of time. I suggest that when less than 10 observations over time are available, the strictly homeostatic model and the nonstationary model be used to derive a "critical range" for assessing future changes. When longer series are available, the more general model might replace the other two for this purpose, if analytical variation has remained reasonably stable (within +/- 20% of its average value) during the period of observation. Much more experience with the use of all three models in health monitoring programs would be highly desirable.