Several kinds of errors - mainly those of "commission" or "omission" are discussed as a basis for setting up an accuracy assessment program. Accuracy itself is defined and the point is made that much depends on just how any class, feature, or material being classified is meaningfully set forth with proper descriptors. Two factors are important in achieving suitable (hopefully, high) accuracy: spatial resolution (which influences the mixed pixel effect) and number of spectral bands involved in the classification. A case study in the Elkton, Maryland area is treated in some detail to demonstrate just what the accuracy of its classification in Landsat data sets really means.

The U.S. Geological Survey has reported results of accuracy assessments of the 1:250,000 and 1:1,000,000 land use maps of Level 1 classifications (see page 4-1), based on aerial photos, that meets the 85% accuracy criterion at the 95% confidence level.

http://fas.org/irp/imint/docs/rst/Sect13/Sect13_3.html