Selectivity estimation is crucial for Query Optimization. Misevaluating the selectivity of the query can result in a poor execution plan choice. 
To aid the task of selectivity estimation, a concise summary of the data distribution is constructed (histograms and samples are the ones most commonly used).  
There has been much work done to provide good estimations. However the estimations are never exact: there is always uncertainty involved. From the point of view 
of Query Optimizer, it is very important to know the degree of uncertainty associated with the estimate. In this presentation we quantify the uncertainty 
involved in selectivity estimations. Specifically, we show that probability distributions and prediction intervals of possible selectivities can be issued 
instead of a point estimates.