Real PCDD and PCDF emission data from a statistically planned measurement programme at Amagerforbrænding, Copenhagen, 1985 are analyzed. It is demonstrated how statistical analysis of variance can be used to break down the total variability in the data into interpretable components. Crossed and nested data structures are considered. Emission differences over time are significant, but less so than differences due to alternative sample taking methods, and even greater differences were found between laboratories.