A number of methods can be applied to characterize the reference condition that is used to assess the biological condition of non-reference sites.  In the predictive modeling approach, reference sites are grouped by their biological similarity to one another. Invariant habitat characteristics that discriminate among groups are then used to predict the group of reference sites that best characterizes the biological potential of the site we want to assess.  This approach involves somewhat arbitrary groupings of sites, and some have argued that a nearest neighbour (NN) approach, which does not involve such groupings of sites, avoids potential problems of this classification step.  We compared traditional biotic groupings and a NN approach using a number of data sets to evaluate the performance (accuracy and precision) of this NN approach for matching test sites with appropriate reference sites.