The performance of an indicator of health or nutritional status depends on its sensitivity and specificity properties over a range of cut-offs. Frequently, it is of interest to compare indicators to pick the best for a given purpose, such as screening for disease or monitoring to detect changes in prevalence of inadequate nutriture. Relative operating characteristic (ROC) analysis provides an objective method for making this comparison, but the application of this methodology as described for epidemiologists in this Journal is now outdated for most indicators. Recent developments are noted and an alternative analysis for use with continuous Gaussian data is presented here. The estimators and statistical test procedures proposed here are compared with the previously described methods, by means of a computer simulation study. The new procedures are found to be superior for continuous Gaussian data, and have the practical advantage that they do not require use of a specialized computer program. The implications of these results for comparing indicators to be used to monitor population prevalences are discussed.