Estimators of the number of additional species expected in the next al
l samples offer a potentially important tool for improving cost-effect
iveness of species inventories but are largely untested. We used Monte
Carlo methods to compare ii such estimators, across a range of commun
ity structures and sampling regimes, and validated our results, where
possible, using empirical data from vascular plant and beetle inventor
ies from Glacier National Park, Montana, USA. We found that B. Efron a
nd R. Thisted's 1976 negative binomial estimator was most robust to di
fferences in community structure and that it was among the most accura
te estimators when sampling was from model communities with structures
resembling the large, heterogeneous communities that are the likely t
argets of major inventory efforts. Other estimators may be preferred u
nder specific conditions, however. For example, when sampling was from
model communities with highly even species-abundance distributions, e
stimates based on the Michaelis-Menten model were most accurate; when
sampling was from moderately even model communities with S = 10 specie
s or communities with highly uneven species-abundance distributions, e
stimates based on Gleason's (1922) species-area model were most accura
te. We suggest that use of such methods in species inventories can hel
p improve cost-effectiveness by providing an objective basis for redir
ecting sampling to more-productive sites, methods, or time periods as
the expectation of detecting additional species becomes unacceptably l
ow.