Empirical analyses of data on working set size are reported. The data do not support the hypothesis that working set sizes are normally distributed. The data suggest various algorithms for predicting working set size based on the program's past history. Several representative algorithms are discussed and evaluated.