The problem of predicting human protein requirements is examined in detail. Three methods from the literature are shown to represent a logical sequence wherin each takes into account more sources of variation than does the previous one. These are the single level method and two versions of the multiple level method, all of which assume a population response curve. A fourth method is proposed as the next step in this sequence. This method calculates a response curve for each individual and works with the individual requirements. Since each method makes fewer assumptions about the data and, therefore, about the biological situation of population protein needs, each method gives successively larger protein allowance values, while at the same time identifying the sources of variability and uncertainty in these estimates.