Statistical analysis of nitrogen balance data is an important approach to the estimation of human nutrient requirement. The usual procedure is to regress nitrogen (N) balance on intake and to define the requirement as intake that would produce zero balance. In the actual application of this methodology, there are a number of options, and in the present study we explore the sensitivity of the regression approach to those options. To put this problem into a realistic context, we examine the current controversy over the lysine requirements of healthy adults. From early N balance studies, investigators concluded that the mean requirement was generally less than 10 mg. kg(-1). d(-1), whereas based on recent (13)C-tracer and metabolic studies, we propose a tentative mean requirement of approximately 30 mg. kg(-1). d(-1). Jones et al. (1956) conducted careful N balance studies from which they derived an estimate of lysine requirement of less than about 8 mg. kg(-1). d(-1). We reanalyzed these data with different choices of modes of analysis, mathematical models, and different assumptions concerning the magnitude of miscellaneous N losses. We find that for these data the choice of a specific mathematical model has only a small effect on resultant estimates of requirement, while estimated requirements are very sensitive to amount of unmeasured losses that are assumed and how the model is applied (whether the aggregate data are fitted in one pass to a single model, or the data for each individual subject are fitted to that individual's unique model). Moreover, our reanalysis suggests that the population requirement for lysine is in the range of 17 to 36 mg. kg(-1). d(-1) and strongly supports a lysine requirement value of about 30 mg. kg(-1). d(-1). In general, our results indicate that whenever possible, N balance data should be analyzed using a square root model fitted to individual data and that the median of the individual requirements so derived be used as the estimate of population requirement. Moreover, clearly any statistical analyses of N balance data should include a sensitivity analysis to determine the influence of underlying assumptions. Finally, the finding that these estimates are highly dependent on the assumed amount of N miscellaneous losses recommends that further studies on these losses and of the factors that influence them are essential.