To evaluate sampling for automated defect inspections, a series of four inspection areas ranging from 4 cm2 were performed. Using an automated patterned wafer inspection system, 20 wafers were inspected for each of the sample sizes. Inspected die were selected in a random pattern across the wafer and remained the same for each inspection. The inspected area within each die was varied to obtain the desired total inspection area. This allowed the data for the different sample sizes to be directly compared and reduced the effect of `defect clustering'. In addition, 10 of the wafers received whole wafer inspection for comparisons to the true defect population parameters. The results were evaluated using statistical techniques. Different sample sizes are compared using statistical process control charts, linear regression, evaluation of alpha and beta risk, and defect Pareto charts. It is concluded that sampling can provide reliable and accurate defect density data for making conclusions about the defect population, if the sampled area is chosen carefully