How should we estimate confidence intervals for a ratio, for example, between counts of colony-forming cells?

When calculating a ratio, of colony counts, for example, experimental hematologists often provide no indication of the uncertainty of the estimates, or the uncertainty is computed using a method with partially unknown properties. We have used empirical data on survival of colony-forming cells after exposure to hydroxyurea to select assumptions regarding the statistical properties of the survival ratios. Then, we have evaluated four alternative methods for constructing confidence intervals for such ratios. With simulations, the adequacy of the methods is shown to depend on the assumed statistical distributions of the data and on the number of observations in the numerator and denominator. Especially with few observations, some methods may produce misleading results. An easily calculable confidence interval, formally based on lognormal distributions, seems to be applicable to a fairly wide range of other plausible statistical distributions as well; it may be a reasonable choice in many situations. Whatever method is chosen, the confidence intervals for ratios are wide, and more data than is generally appreciated are needed for a precise determination of a ratio.