The estimators discussed in the previous section were all defined on an individual flow and to compute a measure (say the number of packets) for a larger aggregate, the analyst would sum the values of the estimators for the flow records matching the aggregate. The sampling decisions for different flows are fortunately independent and thus the variance of the estimates for aggregates are the sum of the respective variances for the estimators for individual flows. In this section we focus on studying the variances of the various estimators for individual flows. We also show that the variances of the estimators based on the core flow slicing algorithm are lower than those of estimators based on random sampling used by Adaptive NetFlow to control memory usage. As in , we start with a simplified setting of a single bin in isolation and then proceed to more realistic settings. The proofs for the variance results from this section can be found in technical report[15].