Size and placement of all features in a mask used for semiconductor integrated circuit manufacture must be within specified ranges for the mask to be acceptable. However the decision on whether the mask meets these specifications is usually based on a limited set of measurements : less than 1 in 10(6) of all possible sites. Conventional variance analysis on such measurements is inadequate and often misleading because mask pattern errors have significant systematic components. Here we describe a more efficient metrology approach for determining whether a mask meets specifications, "adaptive metrology" where the results obtained as the measurements proceed are used to decide how many, and where, more measurements are needed. The process is continued until there is sufficient information to make a decision whether or not the mask is acceptable. In this way the systematic component of the mask errors can be exploited either to increase the accuracy of decisions or to shorten the average measurement time for a given decision accuracy.